agora inbox for [email protected]help / color / mirror / Atom feed
[PATCH v33 11/11] Add documentations about Incremental View Maintenance 126+ messages / 3 participants [nested] [flat]
* [PATCH v33 11/11] Add documentations about Incremental View Maintenance @ 2019-12-20 01:25 Yugo Nagata <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: Yugo Nagata @ 2019-12-20 01:25 UTC (permalink / raw) --- doc/src/sgml/catalogs.sgml | 9 + .../sgml/ref/create_materialized_view.sgml | 124 ++++- .../sgml/ref/refresh_materialized_view.sgml | 8 +- doc/src/sgml/rules.sgml | 437 ++++++++++++++++++ doc/src/sgml/system-views.sgml | 9 + 5 files changed, 583 insertions(+), 4 deletions(-) diff --git a/doc/src/sgml/catalogs.sgml b/doc/src/sgml/catalogs.sgml index b654fae1b2..8ef73edd12 100644 --- a/doc/src/sgml/catalogs.sgml +++ b/doc/src/sgml/catalogs.sgml @@ -2231,6 +2231,15 @@ SCRAM-SHA-256$<replaceable><iteration count></replaceable>:<replaceable>&l </para></entry> </row> + <row> + <entry role="catalog_table_entry"><para role="column_definition"> + <structfield>relisivm</structfield> <type>bool</type> + </para> + <para> + True if relation is incrementally maintainable materialized view + </para></entry> + </row> + <row> <entry role="catalog_table_entry"><para role="column_definition"> <structfield>relrewrite</structfield> <type>oid</type> diff --git a/doc/src/sgml/ref/create_materialized_view.sgml b/doc/src/sgml/ref/create_materialized_view.sgml index 0d2fea2b97..8c574062db 100644 --- a/doc/src/sgml/ref/create_materialized_view.sgml +++ b/doc/src/sgml/ref/create_materialized_view.sgml @@ -21,7 +21,7 @@ PostgreSQL documentation <refsynopsisdiv> <synopsis> -CREATE MATERIALIZED VIEW [ IF NOT EXISTS ] <replaceable>table_name</replaceable> +CREATE [ INCREMENTAL ] MATERIALIZED VIEW [ IF NOT EXISTS ] <replaceable>table_name</replaceable> [ (<replaceable>column_name</replaceable> [, ...] ) ] [ USING <replaceable class="parameter">method</replaceable> ] [ WITH ( <replaceable class="parameter">storage_parameter</replaceable> [= <replaceable class="parameter">value</replaceable>] [, ... ] ) ] @@ -60,6 +60,125 @@ CREATE MATERIALIZED VIEW [ IF NOT EXISTS ] <replaceable>table_name</replaceable> <title>Parameters</title> <variablelist> + <varlistentry> + <term><literal>INCREMENTAL</literal></term> + <listitem> + <para> + If specified, some triggers are automatically created so that the rows + of the materialized view are immediately updated when base tables of the + materialized view are updated. In general, this allows faster update of + the materialized view at a price of slower update of the base tables + because the triggers will be invoked. We call this form of materialized + view as "Incrementally Maintainable Materialized View" (IMMV). + </para> + <para> + When <acronym>IMMV</acronym> is defined without using <command>WITH NO DATA</command>, + a unique index is created on the view automatically if possible. If the view + definition query has a GROUP BY clause, a unique index is created on the columns + of GROUP BY expressions. Also, if the view has DISTINCT clause, a unique index + is created on all columns in the target list. Otherwise, if the view contains all + primary key attritubes of its base tables in the target list, a unique index is + created on these attritubes. In other cases, no index is created. + </para> + <para> + There are restrictions of query definitions allowed to use this + option. The following are supported in query definitions for IMMV: + <itemizedlist> + + <listitem> + <para> + Inner joins (including self-joins). + </para> + </listitem> + + <listitem> + <para> + Some built-in aggregate functions (count, sum, avg, min, max) without a HAVING + clause. + </para> + </listitem> + </itemizedlist> + + Unsupported queries with this option include the following: + + <itemizedlist> + <listitem> + <para> + Outer joins. + </para> + </listitem> + + <listitem> + <para> + Sub-queries. + </para> + </listitem> + + <listitem> + <para> + Aggregate functions other than built-in count, sum, avg, min and max. + </para> + </listitem> + <listitem> + <para> + Aggregate functions with a HAVING clause. + </para> + </listitem> + <listitem> + <para> + DISTINCT ON, WINDOW, VALUES, LIMIT and OFFSET clause. + </para> + </listitem> + </itemizedlist> + + Other restrictions include: + <itemizedlist> + + <listitem> + <para> + IMMVs must be based on simple base tables. It's not supported to + create them on top of views or materialized views. + </para> + </listitem> + + <listitem> + <para> + It is not supported to include system columns in an IMMV. + <programlisting> +CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm02 AS SELECT i,j FROM mv_base_a WHERE xmin = '610'; +ERROR: system column is not supported with IVM + </programlisting> + </para> + </listitem> + + <listitem> + <para> + Non-immutable functions are not supported. + <programlisting> +CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm12 AS SELECT i,j FROM mv_base_a WHERE i = random()::int; +ERROR: functions in IMMV must be marked IMMUTABLE + </programlisting> + </para> + </listitem> + + <listitem> + <para> + IMMVs do not support expressions that contains aggregates + </para> + </listitem> + + <listitem> + <para> + Logical replication does not support IMMVs. + </para> + </listitem> + + </itemizedlist> + + </para> + </listitem> + </varlistentry> + <varlistentry> <term><literal>IF NOT EXISTS</literal></term> <listitem> @@ -155,7 +274,8 @@ CREATE MATERIALIZED VIEW [ IF NOT EXISTS ] <replaceable>table_name</replaceable> This clause specifies whether or not the materialized view should be populated at creation time. If not, the materialized view will be flagged as unscannable and cannot be queried until <command>REFRESH - MATERIALIZED VIEW</command> is used. + MATERIALIZED VIEW</command> is used. Also, if the view is IMMV, + triggers for maintaining the view are not created. </para> </listitem> </varlistentry> diff --git a/doc/src/sgml/ref/refresh_materialized_view.sgml b/doc/src/sgml/ref/refresh_materialized_view.sgml index 8ed43ade80..a4d729bdf0 100644 --- a/doc/src/sgml/ref/refresh_materialized_view.sgml +++ b/doc/src/sgml/ref/refresh_materialized_view.sgml @@ -36,9 +36,13 @@ REFRESH MATERIALIZED VIEW [ CONCURRENTLY ] <replaceable class="parameter">name</ privilege on the materialized view. The old contents are discarded. If <literal>WITH DATA</literal> is specified (or defaults) the backing query is executed to provide the new data, and the materialized view is left in a - scannable state. If <literal>WITH NO DATA</literal> is specified no new + scannable state. If the view is an incrementally maintainable materialized + view (IMMV) and was unpopulated, triggers for maintaining the view are + created. Also, a unique index is created for IMMV if it is possible and the + view doesn't have that yet. + If <literal>WITH NO DATA</literal> is specified no new data is generated and the materialized view is left in an unscannable - state. + state. If the view is IMMV, the triggers are dropped. </para> <para> <literal>CONCURRENTLY</literal> and <literal>WITH NO DATA</literal> may not diff --git a/doc/src/sgml/rules.sgml b/doc/src/sgml/rules.sgml index 7a928bd7b9..73597ea7a5 100644 --- a/doc/src/sgml/rules.sgml +++ b/doc/src/sgml/rules.sgml @@ -1100,6 +1100,443 @@ SELECT word FROM words ORDER BY word <-> 'caterpiler' LIMIT 10; </sect1> +<sect1 id="rules-ivm"> +<title>Incremental View Maintenance</title> + +<indexterm zone="rules-ivm"> + <primary>incremental view maintenance</primary> +</indexterm> + +<indexterm zone="rules-ivm"> + <primary>materialized view</primary> + <secondary>incremental view maintenance</secondary> +</indexterm> + +<indexterm zone="rules-ivm"> + <primary>view</primary> + <secondary>incremental view maintenance</secondary> +</indexterm> + +<sect2 id="rules-ivm-overview"> +<title>Overview</title> + +<para> + Incremental View Maintenance (<acronym>IVM</acronym>) is a way to make + materialized views up-to-date in which only incremental changes are computed + and applied on views rather than recomputing the contents from scratch as + <command>REFRESH MATERIALIZED VIEW</command> does. <acronym>IVM</acronym> + can update materialized views more efficiently than recomputation when only + small parts of the view are changed. +</para> + +<para> + There are two approaches with regard to timing of view maintenance: + immediate and deferred. In immediate maintenance, views are updated in the + same transaction that its base table is modified. In deferred maintenance, + views are updated after the transaction is committed, for example, when the + view is accessed, as a response to user command like <command>REFRESH + MATERIALIZED VIEW</command>, or periodically in background, and so on. + <productname>PostgreSQL</productname> currently implements only a kind of + immediate maintenance, in which materialized views are updated immediately + in AFTER triggers when a base table is modified. +</para> + +<para> + To create materialized views supporting <acronym>IVM</acronym>, use the + <command>CREATE INCREMENTAL MATERIALIZED VIEW</command>, for example: +<programlisting> +CREATE <emphasis>INCREMENTAL</emphasis> MATERIALIZED VIEW mymatview AS SELECT * FROM mytab; +</programlisting> + When a materialized view is created with the <literal>INCREMENTAL</literal> + keyword, some triggers are automatically created so that the view's contents are + immediately updated when its base tables are modified. We call this form + of materialized view an Incrementally Maintainable Materialized View + (<acronym>IMMV</acronym>). +<programlisting> +postgres=# CREATE INCREMENTAL MATERIALIZED VIEW m AS SELECT * FROM t0; +NOTICE: could not create an index on materialized view "m" automatically +HINT: Create an index on the materialized view for effcient incremental maintenance. +SELECT 3 +postgres=# SELECT * FROM m; + i +--- + 1 + 2 + 3 +(3 rows) + +postgres=# INSERT INTO t0 VALUES (4); +INSERT 0 1 +postgres=# SELECT * FROM m; -- automatically updated + i +--- + 1 + 2 + 3 + 4 +(4 rows) +</programlisting> +</para> + +<para> + Some <acronym>IMMV</acronym>s have hidden columns which are added + automatically when a materialized view is created. Their name starts + with <literal>__ivm_</literal> and they contain information required + for maintaining the <acronym>IMMV</acronym>. Such columns are not visible + when the <acronym>IMMV</acronym> is accessed by <literal>SELECT *</literal> + but are visible if the column name is explicitly specified in the target + list. We can also see the hidden columns in <literal>\d</literal> + meta-commands of <command>psql</command> commands. +</para> + +<para> + In general, <acronym>IMMV</acronym>s allow faster updates of materialized + views at the price of slower updates to their base tables. Updates of + <acronym>IMMV</acronym> is slower because triggers will be invoked and the + view is updated in triggers per modification statement. +</para> + +<para> + For example, suppose a normal materialized view defined as below: + +<programlisting> +test=# CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm AS + SELECT a.aid, b.bid, a.abalance, b.bbalance + FROM pgbench_accounts a JOIN pgbench_branches b USING(bid); +SELECT 10000000 + +</programlisting> + + Updating a tuple in a base table of this materialized view is rapid but the + <command>REFRESH MATERIALIZED VIEW</command> command on this view takes a long time: + +<programlisting> +test=# UPDATE pgbench_accounts SET abalance = 1000 WHERE aid = 1; +UPDATE 1 +Time: 0.990 ms + +test=# REFRESH MATERIALIZED VIEW mv_normal ; +REFRESH MATERIALIZED VIEW +Time: 33533.952 ms (00:33.534) +</programlisting> +</para> + +<para> + On the other hand, after creating <acronym>IMMV</acronym> with the same view + definition as below: + +<programlisting> +test=# CREATE INCREMENTAL MATERIALIZED VIEW mv_ivm AS + SELECT a.aid, b.bid, a.abalance, b.bbalance + FROM pgbench_accounts a JOIN pgbench_branches b USING(bid); +test=# UPDATE pgbench_accounts SET abalance = 1000 WHERE aid = 1; +NOTICE: created index "mv_ivm_index" on materialized view "mv_ivm" +</programlisting> + + updating a tuple in a base table takes more than the normal view, + but its content is updated automatically and this is faster than the + <command>REFRESH MATERIALIZED VIEW</command> command. + +<programlisting> +test=# UPDATE pgbench_accounts SET abalance = 1000 WHERE aid = 1; +UPDATE 1 +Time: 13.068 ms +</programlisting> + +</para> + +<para> + Appropriate indexes on <acronym>IMMV</acronym>s are necessary for + efficient <acronym>IVM</acronym> because it looks for tuples to be + updated in <acronym>IMMV</acronym>. If there are no indexes, it + will take a long time. +</para> + +<para> + Therefore, when <acronym>IMMV</acronym> is defined, a unique index is created on the view + automatically if possible. If the view definition query has a GROUP BY clause, a unique + index is created on the columns of GROUP BY expressions. Also, if the view has DISTINCT + clause, a unique index is created on all columns in the target list. Otherwise, if the + view contains all primary key attritubes of its base tables in the target list, a unique + index is created on these attritubes. In other cases, no index is created. +</para> + +<para> + In the previous example, a unique index "mv_ivm_index" is created on aid and bid + columns of materialized view "mv_ivm", and this enables the rapid update of the view. + Dropping this index make updating the view take a loger time. +<programlisting> +test=# DROP INDEX mv_ivm_index; +DROP INDEX +Time: 67.081 ms + +test=# UPDATE pgbench_accounts SET abalance = 1000 WHERE aid = 1; +UPDATE 1 +Time: 16386.245 ms (00:16.386) +</programlisting> + +</para> + +<para> + <acronym>IVM</acronym> is effective when we want to keep a materialized + view up-to-date and small fraction of a base table is modified + infrequently. Due to the overhead of immediate maintenance, <acronym>IVM</acronym> + is not effective when a base table is modified frequently. Also, when a + large part of a base table is modified or large data is inserted into a + base table, <acronym>IVM</acronym> is not effective and the cost of + maintenance can be larger than the <command>REFRESH MATERIALIZED VIEW</command> + command. In such situation, we can use <command>REFRESH MATERIALIZED VIEW</command> + and specify <literal>WITH NO DATA</literal> to disable immediate + maintenance before modifying a base table. After a base table modification, + execute the <command>REFRESH MATERIALIZED VIEW</command> (with <literal>WITH DATA</literal>) + command to refresh the view data and enable immediate maintenance. +</para> + +</sect2> + +<sect2 id="rules-ivm-support"> +<title>Supported View Definitions and Restrictions</title> + +<para> + Currently, we can create <acronym>IMMV</acronym>s using inner joins, and some + aggregates. However, several restrictions apply to the definition of IMMV. +</para> + +<sect3 id="rules-ivm-support-joins"> +<title>Joins</title> +<para> + Inner joins including self-join are supported. Outer joins are not supported. +</para> +</sect3> + +<sect3 id="rules-ivm-support-aggregates"> +<title>Aggregates</title> +<para> + Supported aggregate functions are <function>count</function>, <function>sum</function>, + <function>avg</function>, <function>min</function>, and <function>max</function>. + Currently, only built-in aggregate functions are supported and user defined + aggregates cannot be used. When a base table is modified, the new aggregated + values are incrementally calculated using the old aggregated values and values + of related hidden columns stored in <acronym>IMMV</acronym>. +</para> + +<para> + Note that for <function>min</function> or <function>max</function>, the new values + could be re-calculated from base tables with regard to the affected groups when a + tuple containing the current minimal or maximal values are deleted from a base table. + Therefore, it can takes a long time to update an <acronym>IMMV</acronym> containing + these functions. +</para> + +<para> + Also note that using <function>sum</function> or <function>avg</function> on + <type>real</type> (<type>float4</type>) type or <type>double precision</type> + (<type>float8</type>) type in <acronym>IMMV</acronym> is unsafe. This is + because aggregated values in <acronym>IMMV</acronym> can become different from + results calculated from base tables due to the limited precision of these types. + To avoid this problem, use the <type>numeric</type> type instead. +</para> + + <sect4 id="rules-ivm-restrictions-aggregates"> + <title>Restrictions on Aggregates</title> + <para> + There are the following restrictions: + <itemizedlist> + <listitem> + <para> + If we have a <literal>GROUP BY</literal> clause, expressions specified in + <literal>GROUP BY</literal> must appear in the target list. This is + how tuples to be updated in the <acronym>IMMV</acronym> are identified. + These attributes are used as scan keys for searching tuples in the + <acronym>IMMV</acronym>, so indexes on them are required for efficient + <acronym>IVM</acronym>. + </para> + </listitem> + + <listitem> + <para> + <literal>HAVING</literal> clause cannot be used. + </para> + </listitem> + </itemizedlist> + </para> + </sect4> +</sect3> + +<sect3 id="rules-ivm-general-restricitons"> +<title>Other General Restrictions</title> +<para> + There are other restrictions which generally apply to <acronym>IMMV</acronym>: + <itemizedlist> + <listitem> + <para> + Sub-queries cannot be used. + </para> + </listitem> + + <listitem> + <para> + CTEs cannot be used. + </para> + </listitem> + + <listitem> + <para> + Window functions cannot be used. + </para> + </listitem> + + <listitem> + <para> + <acronym>IMMV</acronym>s must be based on simple base tables. It's not + supported to create them on top of views, materialized views, foreign tables, inhe. + </para> + </listitem> + + <listitem> + <para> + LIMIT and OFFSET clauses cannot be used. + </para> + </listitem> + + <listitem> + <para> + <acronym>IMMV</acronym>s cannot contain system columns. + </para> + </listitem> + + <listitem> + <para> + <acronym>IMMV</acronym>s cannot contain non-immutable functions. + </para> + </listitem> + + <listitem> + <para> + UNION/INTERSECT/EXCEPT clauses cannnot be used. + </para> + </listitem> + + <listitem> + <para> + DISTINCT ON clauses cannot be used. + </para> + </listitem> + + <listitem> + <para> + TABLESAMPLE parameter cannot be used. + </para> + </listitem> + + <listitem> + <para> + inheritance parent tables cannnot be used. + </para> + </listitem> + + <listitem> + <para> + VALUES clause cannnot be used. + </para> + </listitem> + + <listitem> + <para> + <literal>GROUPING SETS</literal> and <literal>FILTER</literal> clauses cannot be used. + </para> + </listitem> + + <listitem> + <para> + FOR UPDATE/SHARE cannot be used. + </para> + </listitem> + + <listitem> + <para> + targetlist cannot contain columns whose name start with <literal>__ivm_</literal>. + </para> + </listitem> + + <listitem> + <para> + targetlist cannot contain expressions which contain an aggregate in it. + </para> + </listitem> + + <listitem> + <para> + Logical replication is not supported, that is, even when a base table + at a publisher node is modified, <acronym>IMMV</acronym>s at subscriber + nodes are not updated. + </para> + </listitem> + + </itemizedlist> +</para> +</sect3> + +</sect2> + +<sect2 id="rules-ivm-distinct"> +<title><literal>DISTINCT</literal></title> + +<para> + <productname>PostgreSQL</productname> supports <acronym>IMMV</acronym> with + <literal>DISTINCT</literal>. For example, suppose a <acronym>IMMV</acronym> + defined with <literal>DISTINCT</literal> on a base table containing duplicate + tuples. When tuples are deleted from the base table, a tuple in the view is + deleted if and only if the multiplicity of the tuple becomes zero. Moreover, + when tuples are inserted into the base table, a tuple is inserted into the + view only if the same tuple doesn't already exist in it. +</para> + +<para> + Physically, an <acronym>IMMV</acronym> defined with <literal>DISTINCT</literal> + contains tuples after eliminating duplicates, and the multiplicity of each tuple + is stored in a hidden column named <literal>__ivm_count__</literal>. +</para> +</sect2> + +<sect2 id="rules-ivm-concurrent-transactions"> +<title>Concurrent Transactions</title> +<para> + Suppose an <acronym>IMMV</acronym> is defined on two base tables and each + table was modified in different a concurrent transaction simultaneously. + In the transaction which was committed first, <acronym>IMMV</acronym> can + be updated considering only the change which happened in this transaction. + On the other hand, in order to update the view correctly in the transaction + which was committed later, we need to know the changes occurred in + both transactions. For this reason, <literal>ExclusiveLock</literal> + is held on an <acronym>IMMV</acronym> immediately after a base table is + modified in <literal>READ COMMITTED</literal> mode to make sure that + the <acronym>IMMV</acronym> is updated in the latter transaction after + the former transaction is committed. In <literal>REPEATABLE READ</literal> + or <literal>SERIALIZABLE</literal> mode, an error is raised immediately + if lock acquisition fails because any changes which occurred in + other transactions are not be visible in these modes and + <acronym>IMMV</acronym> cannot be updated correctly in such situations. + However, as an exception if the view has only one base table and + <command>INSERT</command> is performed on the table, + the lock held on thew view is <literal>RowExclusiveLock</literal>. +</para> +</sect2> + +<sect2 id="rules-ivm-rls"> +<title>Row Level Security</title> +<para> + If some base tables have row level security policy, rows that are not visible + to the materialized view's owner are excluded from the result. In addition, such + rows are excluded as well when views are incrementally maintained. However, if a + new policy is defined or policies are changed after the materialized view was created, + the new policy will not be applied to the view contents. To apply the new policy, + you need to refresh materialized views. +</para> +</sect2> + +</sect1> + <sect1 id="rules-update"> <title>Rules on <command>INSERT</command>, <command>UPDATE</command>, and <command>DELETE</command></title> diff --git a/doc/src/sgml/system-views.sgml b/doc/src/sgml/system-views.sgml index bdc34cf94e..d4a1c99a91 100644 --- a/doc/src/sgml/system-views.sgml +++ b/doc/src/sgml/system-views.sgml @@ -1796,6 +1796,15 @@ SELECT * FROM pg_locks pl LEFT JOIN pg_prepared_xacts ppx </para></entry> </row> + <row> + <entry role="catalog_table_entry"><para role="column_definition"> + <structfield>isimmv</structfield> <type>bool</type> + </para> + <para> + True if materialized view is incrementally maintainable + </para></entry> + </row> + <row> <entry role="catalog_table_entry"><para role="column_definition"> <structfield>definition</structfield> <type>text</type> -- 2.25.1 --Multipart=_Tue__2_Jul_2024_17_03_11_+0900_6OrVBZxOB4o_k6A1 Content-Type: text/x-diff; name="v33-0010-Add-regression-tests-for-Incremental-View-Mainte.patch" Content-Disposition: attachment; filename="v33-0010-Add-regression-tests-for-Incremental-View-Mainte.patch" Content-Transfer-Encoding: 7bit ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH 4/4] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. @ 2021-04-04 00:24 Justin Pryzby <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: Justin Pryzby @ 2021-04-04 00:24 UTC (permalink / raw) ..to help avoid users finding this option and changing it in hopes that it'll make their queries faster, but without reading the documentation or understanding what it does. --- doc/src/sgml/config.sgml | 90 +++++++++---------- src/backend/utils/misc/guc.c | 2 +- src/backend/utils/misc/postgresql.conf.sample | 1 - 3 files changed, 46 insertions(+), 47 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index c4d5126c2a..5c7c5a39ca 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -5788,51 +5788,6 @@ SELECT * FROM parent WHERE key = 2400; </listitem> </varlistentry> - <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> - <term><varname>force_parallel_mode</varname> (<type>enum</type>) - <indexterm> - <primary><varname>force_parallel_mode</varname> configuration parameter</primary> - </indexterm> - </term> - <listitem> - <para> - Allows the use of parallel queries for testing purposes even in cases - where no performance benefit is expected. - The allowed values of <varname>force_parallel_mode</varname> are - <literal>off</literal> (use parallel mode only when it is expected to improve - performance), <literal>on</literal> (force parallel query for all queries - for which it is thought to be safe), and <literal>regress</literal> (like - <literal>on</literal>, but with additional behavior changes as explained - below). - </para> - - <para> - More specifically, setting this value to <literal>on</literal> will add - a <literal>Gather</literal> node to the top of any query plan for which this - appears to be safe, so that the query runs inside of a parallel worker. - Even when a parallel worker is not available or cannot be used, - operations such as starting a subtransaction that would be prohibited - in a parallel query context will be prohibited unless the planner - believes that this will cause the query to fail. If failures or - unexpected results occur when this option is set, some functions used - by the query may need to be marked <literal>PARALLEL UNSAFE</literal> - (or, possibly, <literal>PARALLEL RESTRICTED</literal>). - </para> - - <para> - Setting this value to <literal>regress</literal> has all of the same effects - as setting it to <literal>on</literal> plus some additional effects that are - intended to facilitate automated regression testing. Normally, - messages from a parallel worker include a context line indicating that, - but a setting of <literal>regress</literal> suppresses this line so that the - output is the same as in non-parallel execution. Also, - the <literal>Gather</literal> nodes added to plans by this setting are hidden - in <literal>EXPLAIN</literal> output so that the output matches what - would be obtained if this setting were turned <literal>off</literal>. - </para> - </listitem> - </varlistentry> - <varlistentry id="guc-plan-cache_mode" xreflabel="plan_cache_mode"> <term><varname>plan_cache_mode</varname> (<type>enum</type>) <indexterm> @@ -10231,6 +10186,51 @@ dynamic_library_path = 'C:\tools\postgresql;H:\my_project\lib;$libdir' </listitem> </varlistentry> + <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> + <term><varname>force_parallel_mode</varname> (<type>enum</type>) + <indexterm> + <primary><varname>force_parallel_mode</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + Allows the use of parallel queries for testing purposes even in cases + where no performance benefit is expected. + The allowed values of <varname>force_parallel_mode</varname> are + <literal>off</literal> (use parallel mode only when it is expected to improve + performance), <literal>on</literal> (force parallel query for all queries + for which it is thought to be safe), and <literal>regress</literal> (like + <literal>on</literal>, but with additional behavior changes as explained + below). + </para> + + <para> + More specifically, setting this value to <literal>on</literal> will add + a <literal>Gather</literal> node to the top of any query plan for which this + appears to be safe, so that the query runs inside of a parallel worker. + Even when a parallel worker is not available or cannot be used, + operations such as starting a subtransaction that would be prohibited + in a parallel query context will be prohibited unless the planner + believes that this will cause the query to fail. If failures or + unexpected results occur when this option is set, some functions used + by the query may need to be marked <literal>PARALLEL UNSAFE</literal> + (or, possibly, <literal>PARALLEL RESTRICTED</literal>). + </para> + + <para> + Setting this value to <literal>regress</literal> has all of the same effects + as setting it to <literal>on</literal> plus some additional effects that are + intended to facilitate automated regression testing. Normally, + messages from a parallel worker include a context line indicating that, + but a setting of <literal>regress</literal> suppresses this line so that the + output is the same as in non-parallel execution. Also, + the <literal>Gather</literal> nodes added to plans by this setting are hidden + in <literal>EXPLAIN</literal> output so that the output matches what + would be obtained if this setting were turned <literal>off</literal>. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-ignore-system-indexes" xreflabel="ignore_system_indexes"> <term><varname>ignore_system_indexes</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 27287599ad..06de7c0308 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -4840,7 +4840,7 @@ static struct config_enum ConfigureNamesEnum[] = }, { - {"force_parallel_mode", PGC_USERSET, QUERY_TUNING_OTHER, + {"force_parallel_mode", PGC_USERSET, DEVELOPER_OPTIONS, gettext_noop("Forces use of parallel query facilities."), gettext_noop("If possible, run query using a parallel worker and with parallel restrictions."), GUC_EXPLAIN diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 52c4373a79..50faa50462 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -415,7 +415,6 @@ #from_collapse_limit = 8 #join_collapse_limit = 8 # 1 disables collapsing of explicit # JOIN clauses -#force_parallel_mode = off #jit = on # allow JIT compilation #plan_cache_mode = auto # auto, force_generic_plan or # force_custom_plan -- 2.17.0 --GyRA7555PLgSTuth-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v2 4/4] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. @ 2021-04-04 00:24 Justin Pryzby <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: Justin Pryzby @ 2021-04-04 00:24 UTC (permalink / raw) ..to help avoid users finding this option and changing it in hopes that it'll make their queries faster, but without reading the documentation or understanding what it does. --- doc/src/sgml/config.sgml | 90 +++++++++---------- src/backend/utils/misc/guc.c | 2 +- src/backend/utils/misc/postgresql.conf.sample | 1 - 3 files changed, 46 insertions(+), 47 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index eb154cd669..4568a5c5a0 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -5912,51 +5912,6 @@ SELECT * FROM parent WHERE key = 2400; </listitem> </varlistentry> - <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> - <term><varname>force_parallel_mode</varname> (<type>enum</type>) - <indexterm> - <primary><varname>force_parallel_mode</varname> configuration parameter</primary> - </indexterm> - </term> - <listitem> - <para> - Allows the use of parallel queries for testing purposes even in cases - where no performance benefit is expected. - The allowed values of <varname>force_parallel_mode</varname> are - <literal>off</literal> (use parallel mode only when it is expected to improve - performance), <literal>on</literal> (force parallel query for all queries - for which it is thought to be safe), and <literal>regress</literal> (like - <literal>on</literal>, but with additional behavior changes as explained - below). - </para> - - <para> - More specifically, setting this value to <literal>on</literal> will add - a <literal>Gather</literal> node to the top of any query plan for which this - appears to be safe, so that the query runs inside of a parallel worker. - Even when a parallel worker is not available or cannot be used, - operations such as starting a subtransaction that would be prohibited - in a parallel query context will be prohibited unless the planner - believes that this will cause the query to fail. If failures or - unexpected results occur when this option is set, some functions used - by the query may need to be marked <literal>PARALLEL UNSAFE</literal> - (or, possibly, <literal>PARALLEL RESTRICTED</literal>). - </para> - - <para> - Setting this value to <literal>regress</literal> has all of the same effects - as setting it to <literal>on</literal> plus some additional effects that are - intended to facilitate automated regression testing. Normally, - messages from a parallel worker include a context line indicating that, - but a setting of <literal>regress</literal> suppresses this line so that the - output is the same as in non-parallel execution. Also, - the <literal>Gather</literal> nodes added to plans by this setting are hidden - in <literal>EXPLAIN</literal> output so that the output matches what - would be obtained if this setting were turned <literal>off</literal>. - </para> - </listitem> - </varlistentry> - <varlistentry id="guc-plan-cache_mode" xreflabel="plan_cache_mode"> <term><varname>plan_cache_mode</varname> (<type>enum</type>) <indexterm> @@ -10462,6 +10417,51 @@ dynamic_library_path = 'C:\tools\postgresql;H:\my_project\lib;$libdir' </listitem> </varlistentry> + <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> + <term><varname>force_parallel_mode</varname> (<type>enum</type>) + <indexterm> + <primary><varname>force_parallel_mode</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + Allows the use of parallel queries for testing purposes even in cases + where no performance benefit is expected. + The allowed values of <varname>force_parallel_mode</varname> are + <literal>off</literal> (use parallel mode only when it is expected to improve + performance), <literal>on</literal> (force parallel query for all queries + for which it is thought to be safe), and <literal>regress</literal> (like + <literal>on</literal>, but with additional behavior changes as explained + below). + </para> + + <para> + More specifically, setting this value to <literal>on</literal> will add + a <literal>Gather</literal> node to the top of any query plan for which this + appears to be safe, so that the query runs inside of a parallel worker. + Even when a parallel worker is not available or cannot be used, + operations such as starting a subtransaction that would be prohibited + in a parallel query context will be prohibited unless the planner + believes that this will cause the query to fail. If failures or + unexpected results occur when this option is set, some functions used + by the query may need to be marked <literal>PARALLEL UNSAFE</literal> + (or, possibly, <literal>PARALLEL RESTRICTED</literal>). + </para> + + <para> + Setting this value to <literal>regress</literal> has all of the same effects + as setting it to <literal>on</literal> plus some additional effects that are + intended to facilitate automated regression testing. Normally, + messages from a parallel worker include a context line indicating that, + but a setting of <literal>regress</literal> suppresses this line so that the + output is the same as in non-parallel execution. Also, + the <literal>Gather</literal> nodes added to plans by this setting are hidden + in <literal>EXPLAIN</literal> output so that the output matches what + would be obtained if this setting were turned <literal>off</literal>. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-ignore-system-indexes" xreflabel="ignore_system_indexes"> <term><varname>ignore_system_indexes</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 2b9583cc77..598813da2d 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -4922,7 +4922,7 @@ static struct config_enum ConfigureNamesEnum[] = }, { - {"force_parallel_mode", PGC_USERSET, QUERY_TUNING_OTHER, + {"force_parallel_mode", PGC_USERSET, DEVELOPER_OPTIONS, gettext_noop("Forces use of parallel query facilities."), gettext_noop("If possible, run query using a parallel worker and with parallel restrictions."), GUC_EXPLAIN diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index cc9edc410f..a38fe18886 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -421,7 +421,6 @@ #from_collapse_limit = 8 #join_collapse_limit = 8 # 1 disables collapsing of explicit # JOIN clauses -#force_parallel_mode = off #jit = on # allow JIT compilation #plan_cache_mode = auto # auto, force_generic_plan or # force_custom_plan -- 2.17.0 --0eh6TmSyL6TZE2Uz-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v3 3/3] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. @ 2021-04-04 00:24 Justin Pryzby <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: Justin Pryzby @ 2021-04-04 00:24 UTC (permalink / raw) ..to discourage users from changing this option in hopes that it'll make their queries faster, but without reading the documentation or understanding what it does. --- doc/src/sgml/config.sgml | 96 ++++++++++--------- src/backend/utils/misc/guc.c | 2 +- src/backend/utils/misc/postgresql.conf.sample | 1 - 3 files changed, 50 insertions(+), 49 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index f749fe9ce7..a7350c6ddc 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -5912,51 +5912,6 @@ SELECT * FROM parent WHERE key = 2400; </listitem> </varlistentry> - <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> - <term><varname>force_parallel_mode</varname> (<type>enum</type>) - <indexterm> - <primary><varname>force_parallel_mode</varname> configuration parameter</primary> - </indexterm> - </term> - <listitem> - <para> - Allows the use of parallel queries for testing purposes even in cases - where no performance benefit is expected. - The allowed values of <varname>force_parallel_mode</varname> are - <literal>off</literal> (use parallel mode only when it is expected to improve - performance), <literal>on</literal> (force parallel query for all queries - for which it is thought to be safe), and <literal>regress</literal> (like - <literal>on</literal>, but with additional behavior changes as explained - below). - </para> - - <para> - More specifically, setting this value to <literal>on</literal> will add - a <literal>Gather</literal> node to the top of any query plan for which this - appears to be safe, so that the query runs inside of a parallel worker. - Even when a parallel worker is not available or cannot be used, - operations such as starting a subtransaction that would be prohibited - in a parallel query context will be prohibited unless the planner - believes that this will cause the query to fail. If failures or - unexpected results occur when this option is set, some functions used - by the query may need to be marked <literal>PARALLEL UNSAFE</literal> - (or, possibly, <literal>PARALLEL RESTRICTED</literal>). - </para> - - <para> - Setting this value to <literal>regress</literal> has all of the same effects - as setting it to <literal>on</literal> plus some additional effects that are - intended to facilitate automated regression testing. Normally, - messages from a parallel worker include a context line indicating that, - but a setting of <literal>regress</literal> suppresses this line so that the - output is the same as in non-parallel execution. Also, - the <literal>Gather</literal> nodes added to plans by this setting are hidden - in <literal>EXPLAIN</literal> output so that the output matches what - would be obtained if this setting were turned <literal>off</literal>. - </para> - </listitem> - </varlistentry> - <varlistentry id="guc-plan-cache_mode" xreflabel="plan_cache_mode"> <term><varname>plan_cache_mode</varname> (<type>enum</type>) <indexterm> @@ -10374,8 +10329,10 @@ dynamic_library_path = 'C:\tools\postgresql;H:\my_project\lib;$libdir' <title>Developer Options</title> <para> - The following parameters are intended for work on the - <productname>PostgreSQL</productname> source code, and in some cases + The following parameters are intended for development with + <productname>PostgreSQL</productname>. Some of them are useful while + writing SQL queries, some of them are useful when + working on the source code of PostgreSQL itself, and to assist with recovery of severely damaged databases. There should be no reason to use them on a production database. As such, they have been excluded from the sample @@ -10464,6 +10421,51 @@ dynamic_library_path = 'C:\tools\postgresql;H:\my_project\lib;$libdir' </listitem> </varlistentry> + <varlistentry id="guc-force-parallel-mode" xreflabel="force_parallel_mode"> + <term><varname>force_parallel_mode</varname> (<type>enum</type>) + <indexterm> + <primary><varname>force_parallel_mode</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + Allows the use of parallel queries for testing purposes even in cases + where no performance benefit is expected. + The allowed values of <varname>force_parallel_mode</varname> are + <literal>off</literal> (use parallel mode only when it is expected to improve + performance), <literal>on</literal> (force parallel query for all queries + for which it is thought to be safe), and <literal>regress</literal> (like + <literal>on</literal>, but with additional behavior changes as explained + below). + </para> + + <para> + More specifically, setting this value to <literal>on</literal> will add + a <literal>Gather</literal> node to the top of any query plan for which this + appears to be safe, so that the query runs inside of a parallel worker. + Even when a parallel worker is not available or cannot be used, + operations such as starting a subtransaction that would be prohibited + in a parallel query context will be prohibited unless the planner + believes that this will cause the query to fail. If failures or + unexpected results occur when this option is set, some functions used + by the query may need to be marked <literal>PARALLEL UNSAFE</literal> + (or, possibly, <literal>PARALLEL RESTRICTED</literal>). + </para> + + <para> + Setting this value to <literal>regress</literal> has all of the same effects + as setting it to <literal>on</literal> plus some additional effects that are + intended to facilitate automated regression testing. Normally, + messages from a parallel worker include a context line indicating that, + but a setting of <literal>regress</literal> suppresses this line so that the + output is the same as in non-parallel execution. Also, + the <literal>Gather</literal> nodes added to plans by this setting are hidden + in <literal>EXPLAIN</literal> output so that the output matches what + would be obtained if this setting were turned <literal>off</literal>. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-ignore-system-indexes" xreflabel="ignore_system_indexes"> <term><varname>ignore_system_indexes</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index a3c2ebbc53..b8738561e6 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -4922,7 +4922,7 @@ static struct config_enum ConfigureNamesEnum[] = }, { - {"force_parallel_mode", PGC_USERSET, QUERY_TUNING_OTHER, + {"force_parallel_mode", PGC_USERSET, DEVELOPER_OPTIONS, gettext_noop("Forces use of parallel query facilities."), gettext_noop("If possible, run query using a parallel worker and with parallel restrictions."), GUC_EXPLAIN diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 2f6dd014a8..0f7f49b949 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -421,7 +421,6 @@ #from_collapse_limit = 8 #join_collapse_limit = 8 # 1 disables collapsing of explicit # JOIN clauses -#force_parallel_mode = off #jit = on # allow JIT compilation #plan_cache_mode = auto # auto, force_generic_plan or # force_custom_plan -- 2.17.0 --oC1+HKm2/end4ao3-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
* [PATCH v8] using indexscan to speedup add not null constraints @ 2025-10-20 05:29 jian he <[email protected]> 0 siblings, 0 replies; 126+ messages in thread From: jian he @ 2025-10-20 05:29 UTC (permalink / raw) This patch tries to use index_beginscan() / index_getnext() / index_endscan() mentioned in [1] to speedup adding not-null constraints to the existing table. The main logic happens in phase3 ATRewriteTable 1. collect all not-null constraints. 2. For each not-null constraint, check whether there is a corresponding index available for validation. If not, then can not use indexscan to verify not-null constraints. 3. If any of the following conditions are true * table scan or rewrite is required, * table wasn't locked with `AccessExclusiveLock` * the NOT NULL constraint applies to a virtual generated column then index scan cannot be used for fast validation. 4. If all conditions are satisfied, attempt to use indexscan to verify whether the column contains any NULL values. concurrency concern: ALTER TABLE SET NOT NULL will take an ACCESS EXCLUSIVE lock, so there is less variant of racing issue can occur? to prove accurate, I wrote some isolation tests. see[2] performance: demo: case when: %20 percent values are NULL and have been deleted from heap but they still on the index. drop table if exists t; create unlogged table t(a int, b int) with (autovacuum_enabled = off, vacuum_index_cleanup=off); insert into t select case when g % 5 = 0 then null else g end, g+1 from generate_series(1,1_000_000) g; create index t_idx_a on t(a); delete from t where a is null; alter table t alter column a drop not null; alter table t add constraint t1 not null a; the above two statement running several times: patch Time:: 1.084 ms master Time: 12.045 ms references: [1] https://postgr.es/m/CA%2BTgmoa5NKz8iGW_9v7wz%3D-%2BzQFu%3DE4SZoaTaU1znLaEXRYp-Q%40mail.gmail.com [2] https://postgr.es/m/900056D1-32DF-4927-8251-3E0C0DC407FD%40anarazel.de discussion: https://postgr.es/m/CACJufxFiW=4k1is=F1J=r-Cx1RuByXQPUrWB331U47rSnGz+hw@mail.gmail.com commitfest entry: https://commitfest.postgresql.org/patch/5444 --- src/backend/commands/tablecmds.c | 116 +++++++++- src/backend/executor/execIndexing.c | 213 ++++++++++++++++++ src/include/executor/executor.h | 2 + .../expected/indexscan-check-notnull.out | 97 ++++++++ src/test/isolation/isolation_schedule | 1 + .../specs/indexscan-check-notnull.spec | 45 ++++ src/test/modules/test_misc/meson.build | 1 + .../t/011_indexscan_validate_notnull.pl | 163 ++++++++++++++ 8 files changed, 632 insertions(+), 6 deletions(-) create mode 100644 src/test/isolation/expected/indexscan-check-notnull.out create mode 100644 src/test/isolation/specs/indexscan-check-notnull.spec create mode 100644 src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index f976c0e5c7e..2af06630b35 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -210,16 +210,19 @@ typedef struct AlteredTableInfo List *changedStatisticsDefs; /* string definitions of same */ } AlteredTableInfo; -/* Struct describing one new constraint to check in Phase 3 scan */ -/* Note: new not-null constraints are handled elsewhere */ +/* + * Struct describing one new constraint to check in Phase 3 scan. Note: new + * not-null constraints are handled here too. +*/ typedef struct NewConstraint { char *name; /* Constraint name, or NULL if none */ - ConstrType contype; /* CHECK or FOREIGN */ + ConstrType contype; /* CHECK or FOREIGN or NOT NULL */ Oid refrelid; /* PK rel, if FOREIGN */ Oid refindid; /* OID of PK's index, if FOREIGN */ bool conwithperiod; /* Whether the new FOREIGN KEY uses PERIOD */ Oid conid; /* OID of pg_constraint entry, if FOREIGN */ + int attnum; /* NOT NULL constraint attribute number */ Node *qual; /* Check expr or CONSTR_FOREIGN Constraint */ ExprState *qualstate; /* Execution state for CHECK expr */ } NewConstraint; @@ -6217,6 +6220,9 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) needscan = true; con->qualstate = ExecPrepareExpr((Expr *) expand_generated_columns_in_expr(con->qual, oldrel, 1), estate); break; + case CONSTR_NOTNULL: + /* Nothing to do here */ + break; case CONSTR_FOREIGN: /* Nothing to do here */ break; @@ -6270,10 +6276,69 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap) wholeatt->attnum); } } - if (notnull_attrs || notnull_virtual_attrs) - needscan = true; } + /*- + * If we have to verify new not-null constraints and don't have any other + * reason to fully scan the table, then we can use indexes to test for + * absence of NULL values instead of scanning. The following conditions + * must be met: + * - we're operating on a regular table + * - we hold AccessExclusiveLock on the table + * - no not-null constraint covers a virtual generated column, + * because no indexes can be created on them + */ + if (!needscan && + newrel == NULL && + tab->verify_new_notnull && + notnull_virtual_attrs == NIL && + oldrel->rd_rel->relkind == RELKIND_RELATION && + CheckRelationLockedByMe(oldrel, AccessExclusiveLock, false)) + { + List *notnull_attnums = NIL; + + Assert(!tab->rewrite); + + foreach(l, tab->constraints) + { + Form_pg_attribute attr; + NewConstraint *con = lfirst(l); + + if (con->contype != CONSTR_NOTNULL) + continue; + + attr = TupleDescAttr(newTupDesc, con->attnum - 1); + + Assert(attr->attnotnull); + Assert(attr->attnum == con->attnum); + + if (attr->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL) + { + needscan = true; + break; + } + + notnull_attnums = lappend_int(notnull_attnums, attr->attnum); + } + + /* + * Validate new NOT NULL constraints, unless we need a scan for other + * reasons, because we can validate them cheaply there, and doing it + * also here would be a waste of cycles. + */ + if (notnull_attnums != NIL && !needscan) + { + if (!index_check_notnull(oldrel, notnull_attnums)) + needscan = true; + else + ereport(DEBUG1, + errmsg_internal("all new not-null constraints on relation \"%s\" have been validated by using index scan", + RelationGetRelationName(oldrel))); + } + } + else if (notnull_attrs || notnull_virtual_attrs) + needscan = true; + if (newrel || needscan) { ExprContext *econtext; @@ -7949,6 +8014,7 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, CookedConstraint *ccon; List *cooked; bool is_no_inherit = false; + AlteredTableInfo *tab = NULL; /* Guard against stack overflow due to overly deep inheritance tree. */ check_stack_depth(); @@ -8073,10 +8139,32 @@ ATExecSetNotNull(List **wqueue, Relation rel, char *conName, char *colName, constraint->is_no_inherit = is_no_inherit; constraint->conname = conName; + tab = ATGetQueueEntry(wqueue, rel); + /* and do it */ cooked = AddRelationNewConstraints(rel, NIL, list_make1(constraint), false, !recursing, false, NULL); ccon = linitial(cooked); + + Assert(ccon->contype == CONSTR_NOTNULL); + + /* + * We may use indexscan to vertify not-null constraint in Phase3. Add the + * to-be-validated not-null constraint to Phase 3's queue. + */ + if (!ccon->skip_validation) + { + NewConstraint *newcon; + + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + + newcon->name = ccon->name; + newcon->contype = ccon->contype; + newcon->attnum = ccon->attnum; + + tab->constraints = lappend(tab->constraints, newcon); + } + ObjectAddressSet(address, ConstraintRelationId, ccon->conoid); InvokeObjectPostAlterHook(RelationRelationId, @@ -9983,13 +10071,15 @@ ATAddCheckNNConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel, { CookedConstraint *ccon = (CookedConstraint *) lfirst(lcon); - if (!ccon->skip_validation && ccon->contype != CONSTR_NOTNULL) + if (!ccon->skip_validation) { NewConstraint *newcon; newcon = palloc0_object(NewConstraint); newcon->name = ccon->name; newcon->contype = ccon->contype; + if (ccon->contype == CONSTR_NOTNULL) + newcon->attnum = ccon->attnum; newcon->qual = ccon->expr; tab->constraints = lappend(tab->constraints, newcon); @@ -13258,6 +13348,7 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, List *children = NIL; AttrNumber attnum; char *colname; + NewConstraint *newcon; con = (Form_pg_constraint) GETSTRUCT(contuple); Assert(con->contype == CONSTRAINT_NOTNULL); @@ -13321,7 +13412,20 @@ QueueNNConstraintValidation(List **wqueue, Relation conrel, Relation rel, /* Set attnotnull appropriately without queueing another validation */ set_attnotnull(NULL, rel, attnum, true, false); + /* + * Queue validation for phase 3. + * + * ALTER TABLE SET NOT NULL will add NOT NULL constraint to + * AlteredTableInfo->constraints, for consistency, do the same here. + */ + newcon = (NewConstraint *) palloc0(sizeof(NewConstraint)); + newcon->name = colname; + newcon->contype = CONSTR_NOTNULL; + newcon->attnum = attnum; + + /* Find or create work queue entry for this table */ tab = ATGetQueueEntry(wqueue, rel); + tab->constraints = lappend(tab->constraints, newcon); tab->verify_new_notnull = true; /* diff --git a/src/backend/executor/execIndexing.c b/src/backend/executor/execIndexing.c index 6ae0f959592..88694ce0335 100644 --- a/src/backend/executor/execIndexing.c +++ b/src/backend/executor/execIndexing.c @@ -108,12 +108,15 @@ #include "access/genam.h" #include "access/relscan.h" +#include "access/table.h" #include "access/tableam.h" #include "access/xact.h" #include "catalog/index.h" +#include "catalog/pg_am_d.h" #include "executor/executor.h" #include "nodes/nodeFuncs.h" #include "storage/lmgr.h" +#include "utils/fmgroids.h" #include "utils/injection_point.h" #include "utils/multirangetypes.h" #include "utils/rangetypes.h" @@ -146,6 +149,7 @@ static bool index_expression_changed_walker(Node *node, Bitmapset *allUpdatedCols); static void ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char typtype, Oid atttypid); +static bool index_check_notnull_internal(Relation relation, List *attnums, List *idxs); /* ---------------------------------------------------------------- * ExecOpenIndices @@ -1178,3 +1182,212 @@ ExecWithoutOverlapsNotEmpty(Relation rel, NameData attname, Datum attval, char t errmsg("empty WITHOUT OVERLAPS value found in column \"%s\" in relation \"%s\"", NameStr(attname), RelationGetRelationName(rel)))); } + +/* + * notnull_attnums: list of attribute numbers of all new not-null constraints. + * + * First check whether the relation has a suitable index for each not-null + * attribute, then using indexscan mechanism to verify that *all* attributes on + * notnull_attnums are indeed not null. + * + * Returning true means all new not-null constraints have been verified, no + * need an extra scan to verify newly added not-null constraints. + */ +bool +index_check_notnull(Relation relation, List *notnull_attnums) +{ + Relation pg_index; + Relation index_rel; + SysScanDesc indscan; + ScanKeyData skey; + HeapTuple htup; + TupleDesc tupdesc; + Form_pg_attribute attr; + List *idxs = NIL; + List *attnums = NIL; + + tupdesc = RelationGetDescr(relation); + + /* Prepare to scan pg_index for entries having indrelid = this rel */ + ScanKeyInit(&skey, + Anum_pg_index_indrelid, + BTEqualStrategyNumber, F_OIDEQ, + ObjectIdGetDatum(RelationGetRelid(relation))); + + pg_index = table_open(IndexRelationId, AccessShareLock); + indscan = systable_beginscan(pg_index, IndexIndrelidIndexId, true, + NULL, 1, &skey); + + while (HeapTupleIsValid(htup = systable_getnext(indscan))) + { + Form_pg_index index = (Form_pg_index) GETSTRUCT(htup); + + /* + * we can only use non-deferred, validated, and alive btree index to + * validate not-null constraints + */ + if (!index->indimmediate || !index->indisvalid || !index->indislive) + continue; + + /* can not use expression index or partial index too */ + if (!heap_attisnull(htup, Anum_pg_index_indexprs, NULL) || + !heap_attisnull(htup, Anum_pg_index_indpred, NULL)) + continue; + + index_rel = index_open(index->indexrelid, AccessShareLock); + + /* can only apply to btree index */ + if (index_rel->rd_rel->relam != BTREE_AM_OID) + { + index_close(index_rel, AccessShareLock); + continue; + } + + /* collect index attnums while loop */ + for (int i = 0; i < index->indnkeyatts; i++) + { + attr = TupleDescAttr(tupdesc, (index->indkey.values[i] - 1)); + + if (attr->attisdropped) + continue; + + if (list_member_int(notnull_attnums, attr->attnum) && + !list_member_int(attnums, attr->attnum)) + { + attnums = lappend_int(attnums, attr->attnum); + idxs = lappend_oid(idxs, index->indexrelid); + } + } + index_close(index_rel, NoLock); + } + systable_endscan(indscan); + table_close(pg_index, AccessShareLock); + + if (attnums == NIL) + return false; + + /* + * every not-null constraint requires an index for indexscan. If no such + * index exists, we need to use regular table scan to validate not-null + * constraints. + */ + foreach_int(attno, notnull_attnums) + { + if (!list_member_int(attnums, attno)) + return false; + } + + return index_check_notnull_internal(relation, attnums, idxs); +} + +/* + * Use indexscan mechanism to fast verify each attribute in "attnums" are + * not-null or not. + */ +static bool +index_check_notnull_internal(Relation relation, List *attnums, List *idxs) +{ + EState *estate; + ExprContext *econtext; + TupleTableSlot *existing_slot; + bool all_not_null = true; + ListCell *lc, + *lc2; + + /* + * Need an EState for slot to hold the current tuple. + */ + estate = CreateExecutorState(); + econtext = GetPerTupleExprContext(estate); + + forboth(lc, attnums, lc2, idxs) + { + SnapshotData DirtySnapshot; + IndexScanDesc index_scan; + ScanKeyData scankeys[INDEX_MAX_KEYS]; + IndexInfo *indexInfo; + AttrNumber sk_attno = -1; + Relation index; + int indnkeyatts; + + AttrNumber attno = lfirst_int(lc); + Oid indexoid = lfirst_oid(lc2); + + existing_slot = table_slot_create(relation, NULL); + + /* Arrange for econtext's scan tuple to be the tuple under test */ + econtext->ecxt_scantuple = existing_slot; + + index = index_open(indexoid, NoLock); + + indexInfo = BuildIndexInfo(index); + indnkeyatts = IndexRelationGetNumberOfKeyAttributes(index); + + /* + * Search the tuples that are in the index for any violations, + * including tuples that aren't visible yet. + */ + InitDirtySnapshot(DirtySnapshot); + + for (int i = 0; i < indnkeyatts; i++) + { + if (indexInfo->ii_IndexAttrNumbers[i] == attno) + { + sk_attno = i + 1; + break; + } + } + + if (sk_attno == -1) + elog(ERROR, "cache lookup failed for attribute number %d on index %u", indexoid, attno); + + for (int i = 0; i < indnkeyatts; i++) + { + /* set up an IS NULL scan key so that we ignore not nulls */ + ScanKeyEntryInitialize(&scankeys[i], + SK_ISNULL | SK_SEARCHNULL, + sk_attno, /* index col to scan */ + InvalidStrategy, /* no strategy */ + InvalidOid, /* no strategy subtype */ + InvalidOid, /* no collation */ + InvalidOid, /* no reg proc for this */ + (Datum) 0); /* constant */ + } + + index_scan = index_beginscan(relation, index, &DirtySnapshot, NULL, indnkeyatts, 0); + index_rescan(index_scan, scankeys, indnkeyatts, NULL, 0); + + while (index_getnext_slot(index_scan, ForwardScanDirection, existing_slot)) + { + Datum values[INDEX_MAX_KEYS]; + bool nulls[INDEX_MAX_KEYS]; + + /* + * Extract the index column values and isnull flags from the + * existing tuple. + */ + FormIndexDatum(indexInfo, existing_slot, estate, values, nulls); + + if (nulls[sk_attno - 1]) + { + all_not_null = false; + break; + } + } + + index_endscan(index_scan); + index_close(index, NoLock); + ExecDropSingleTupleTableSlot(existing_slot); + + /* exit earlier */ + if (!all_not_null) + { + FreeExecutorState(estate); + return false; + } + } + + FreeExecutorState(estate); + + return true; +} diff --git a/src/include/executor/executor.h b/src/include/executor/executor.h index 5929aabc353..7f24caeeafb 100644 --- a/src/include/executor/executor.h +++ b/src/include/executor/executor.h @@ -756,6 +756,8 @@ extern void check_exclusion_constraint(Relation heap, Relation index, const Datum *values, const bool *isnull, EState *estate, bool newIndex); +extern bool index_check_notnull(Relation relation, List *notnull_attnums); + /* * prototypes from functions in execReplication.c */ diff --git a/src/test/isolation/expected/indexscan-check-notnull.out b/src/test/isolation/expected/indexscan-check-notnull.out new file mode 100644 index 00000000000..c6a652e2518 --- /dev/null +++ b/src/test/isolation/expected/indexscan-check-notnull.out @@ -0,0 +1,97 @@ +Parsed test spec with 2 sessions + +starting permutation: s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s1_d1: DELETE FROM t; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; <waiting ...> +step s1_c1: COMMIT; +step s2_a1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + + +starting permutation: s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +ERROR: column "a" of relation "t" contains null values +step s1_r1: ROLLBACK; +step s2_r1: ROLLBACK; + +starting permutation: s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +step s1_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s2_b2: BEGIN; +step s2_d1: DROP INDEX t_ab_idx; +step s1_d1: DELETE FROM t; <waiting ...> +step s2_r1: ROLLBACK; +step s1_d1: <... completed> +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + +step s1_r1: ROLLBACK; + +starting permutation: s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 +step s1_b2: BEGIN; +step s2_b1: BEGIN ISOLATION LEVEL REPEATABLE READ; +step s1_d1: DELETE FROM t; +step s2_d1: DROP INDEX t_ab_idx; <waiting ...> +step s1_s0: SAVEPOINT s0; +step s1_a1: ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a; +step s1_c1: COMMIT; +step s2_d1: <... completed> +step s2_c1: COMMIT; +step s1_q1: SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + +conname|conrelid|contype|convalidated +-------+--------+-------+------------ +t1_nn |t |n |t +(1 row) + diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule index 01ff1c6586f..7ce6bbcaf89 100644 --- a/src/test/isolation/isolation_schedule +++ b/src/test/isolation/isolation_schedule @@ -58,6 +58,7 @@ test: merge-delete test: merge-update test: merge-match-recheck test: merge-join +test: indexscan-check-notnull test: delete-abort-savept test: delete-abort-savept-2 test: aborted-keyrevoke diff --git a/src/test/isolation/specs/indexscan-check-notnull.spec b/src/test/isolation/specs/indexscan-check-notnull.spec new file mode 100644 index 00000000000..05b4488550c --- /dev/null +++ b/src/test/isolation/specs/indexscan-check-notnull.spec @@ -0,0 +1,45 @@ + +# +# using indexscan to check a column not-null constraint +# is satisfied or not. +# + +setup +{ + CREATE TABLE t (a int, b int); + CREATE INDEX t_ab_idx on t(a,b); + INSERT INTO t values (null, 1); + INSERT INTO t SELECT x, x*10 FROM generate_series(1,3) g(x); +} + +teardown +{ + DROP TABLE t; +} + +session s1 +step s1_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s1_b2 { BEGIN; } +step s1_d1 { DELETE FROM t;} +step s1_s0 { SAVEPOINT s0;} +step s1_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s1_q1 { SELECT conname, conrelid::regclass, contype, convalidated + FROM pg_constraint WHERE conname = 't1_nn'; + } +step s1_r1 { ROLLBACK; } +step s1_c1 { COMMIT; } + +session s2 +step s2_b1 { BEGIN ISOLATION LEVEL REPEATABLE READ; } +step s2_b2 { BEGIN; } +step s2_a1 { ALTER TABLE t ADD CONSTRAINT t1_nn NOT NULL a;} +step s2_d1 { DROP INDEX t_ab_idx;} +step s2_r1 { ROLLBACK; } +step s2_c1 { COMMIT; } + +permutation s1_b1 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b2 s2_b1 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b2 s1_d1 s2_a1 s1_c1 s2_c1 s1_q1 +permutation s1_b1 s2_b1 s2_a1 s1_r1 s2_r1 +permutation s1_b1 s2_b2 s2_d1 s1_d1 s2_r1 s1_a1 s1_q1 s1_r1 +permutation s1_b2 s2_b1 s1_d1 s2_d1 s1_s0 s1_a1 s1_c1 s2_c1 s1_q1 diff --git a/src/test/modules/test_misc/meson.build b/src/test/modules/test_misc/meson.build index 6e8db1621a7..56faf359b29 100644 --- a/src/test/modules/test_misc/meson.build +++ b/src/test/modules/test_misc/meson.build @@ -19,6 +19,7 @@ tests += { 't/008_replslot_single_user.pl', 't/009_log_temp_files.pl', 't/010_index_concurrently_upsert.pl', + 't/011_indexscan_validate_notnull.pl', ], # The injection points are cluster-wide, so disable installcheck 'runningcheck': false, diff --git a/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl new file mode 100644 index 00000000000..a3f3cb409b6 --- /dev/null +++ b/src/test/modules/test_misc/t/011_indexscan_validate_notnull.pl @@ -0,0 +1,163 @@ +# Copyright (c) 2025, PostgreSQL Global Development Group + +# Verify that indexscan mechanism can speedup ALTER TABLE ADD NOT NULL + + +use strict; +use warnings FATAL => 'all'; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Initialize a test cluster +my $node = PostgreSQL::Test::Cluster->new('primary'); +$node->init(); +# Turn message level up to DEBUG1 so that we get the messages we want to see +$node->append_conf('postgresql.conf', 'client_min_messages = DEBUG1'); +$node->start; + +# Run a SQL command and return psql's stderr (including debug messages) +sub run_sql_command +{ + my $sql = shift; + my $stderr; + + $node->psql( + 'postgres', + $sql, + stderr => \$stderr, + on_error_die => 1, + on_error_stop => 1); + return $stderr; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_table_verified +{ + my $output = shift; + return index($output, 'DEBUG: verifying table') != -1; +} + +# Check whether result of run_sql_command shows that we did a verify pass +sub is_indexscan_veritify_notnull +{ + my $output = shift; + return index($output, 'DEBUG: all new not-null constraints on relation') != -1; +} + +my $output; + +note "test alter table SET NOT NULL using indexscan with partitioned table"; + +run_sql_command( + 'CREATE TABLE tp (a int, b int, c int, d int ) PARTITION BY range(a); + CREATE TABLE tpp1 partition of tp for values from ( 1 ) to (10); + CREATE TABLE tpp2 partition of tp for values from ( 10 ) to (21); + INSERT INTO tp select g, g + 10, g + 15, g + random(min=>1::int, max=>10::int) from generate_series(1,19) g; + CREATE INDEX ON tp(b); '); + +$output = run_sql_command('ALTER TABLE tp ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), + 'column b will use indexscan to verify not-null status, no need scan table'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# use indexes on partitions to fast vertifing not-null should be ok +$output = run_sql_command('ALTER TABLE tpp1 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null will not scan table'); +ok(!is_table_verified($output), 'no table scan for pp1'); +ok( $output =~ + m/all new not-null constraints on relation "tpp1" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +$output = run_sql_command('ALTER TABLE tpp2 ALTER COLUMN b SET NOT NULL;'); +ok(is_indexscan_veritify_notnull($output), 'use indexscan to validate column b not-null, will not scan table'); +ok(!is_table_verified($output), 'no table scan for tpp2'); +ok( $output =~ + m/all new not-null constraints on relation "tpp2" have been validated by using index scan/, + 'all newly added constraints proved by indexscan'); + +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL;'); + +# table rewrite then can not use indexscan to verify not-null constraint +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b SET NOT NULL, ALTER COLUMN b SET DATA TYPE bigint;' +); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'table tp will rewrite, can not use indexscan to verify not-null constraints on it'); + +# test using indexscan with multiple not-null constraints +run_sql_command( + 'ALTER TABLE tp ALTER COLUMN b DROP NOT NULL; + CREATE INDEX ON tp(c); + CREATE INDEX ON tp(d); '); + +# one column can use indexscan, another can not, so can not use indexscan to verify not-null +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN a SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c'); +ok(is_table_verified($output), 'table scan for tp'); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to verify not-null constraints on table tp'); + +# all columns have index on it +# use indexscan to verify multiple not-null constraints should just fine +$output = run_sql_command( + 'ALTER TABLE tp ALTER COLUMN d SET NOT NULL, ADD CONSTRAINT c_nn NOT NULL c, ALTER COLUMN b SET NOT NULL;'); +ok(!is_table_verified($output), 'no table scan for tp'); +ok(is_indexscan_veritify_notnull($output), + 'using indexscan to verify not-null constraints on column b, c, d, no need table scan'); + +# test with non-partitioned table and other non-Btree index. +run_sql_command( + 'CREATE TABLE t1(f1 INT, f2 int, f3 int,f4 int, f5 int); + INSERT INTO t1 select g, g+1, g+2, g+3, g+4 from generate_series(1, 100) g; + CREATE INDEX t1_f1_f2_idx ON t1(f1,f2); + CREATE UNIQUE INDEX t1_f3idx ON t1(f3); + CREATE INDEX t1_f3f4idx ON t1(f3) include(f4); + CREATE INDEX hash_f5_idx ON t1 USING hash (f5 int4_ops); '); + +# ALTER TABLE VALIDATE CONSTRAINT can not use indexscan, +# because it only hold ShareUpdateExclusiveLock lock. +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn_t1_f2 NOT NULL f2 NOT VALID;'); + +$output = run_sql_command( + 'ALTER TABLE t1 VALIDATE CONSTRAINT nn_t1_f2;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan for ALTER TABLE VALIDATE CONSTRAINT command'); +run_sql_command('ALTER TABLE t1 DROP CONSTRAINT nn_t1_f2;'); + +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f1 SET NOT NULL, ALTER f1 type bigint;' +); +ok(!is_indexscan_veritify_notnull($output), 'table t1 will rewrite, can not use indexscan to validate not-null constraints on t1'); + +run_sql_command('ALTER TABLE t1 ALTER COLUMN f1 DROP NOT NULL;'); +run_sql_command('ALTER TABLE t1 ADD CONSTRAINT nn NOT NULL f1, ALTER f1 type bigint;'); + +# can use indexscan mechanism to fast add multiple not-null constraints +$output = run_sql_command( + 'ALTER TABLE t1 ALTER COLUMN f2 SET NOT NULL, ALTER COLUMN f1 SET NOT NULL;' +); +ok( $output =~ + m/all new not-null constraints on relation "t1" have been validated by using index scan/, + 'all not-null constraints are validated by indexscan'); + +# using indexscan mechanism to fast add a not-null can only apply to key columns, not include column +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nnf4 NOT NULL f4;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraint on t1'); + +# if index is not btree index then cannot use indexscan mechanism to fast add not-null constraint +$output = run_sql_command( + 'ALTER TABLE t1 ADD CONSTRAINT nn_f5 NOT NULL f5;' +); +ok(!is_indexscan_veritify_notnull($output), 'can not use indexscan to validate not-null constraints on t1'); + +$node->stop('fast'); + +done_testing(); -- 2.47.3 --xpkr6zhfzd3fjt76-- ^ permalink raw reply [nested|flat] 126+ messages in thread
end of thread, other threads:[~2025-10-20 05:29 UTC | newest] Thread overview: 126+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2019-12-20 01:25 [PATCH v33 11/11] Add documentations about Incremental View Maintenance Yugo Nagata <[email protected]> 2021-04-04 00:24 [PATCH v2 4/4] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. Justin Pryzby <[email protected]> 2021-04-04 00:24 [PATCH v3 3/3] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. Justin Pryzby <[email protected]> 2021-04-04 00:24 [PATCH 4/4] Change force_parallel_mode to a DEVELOPER GUC, and remove it from sample config.. Justin Pryzby <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]> 2025-10-20 05:29 [PATCH v8] using indexscan to speedup add not null constraints jian he <[email protected]>
This inbox is served by agora; see mirroring instructions for how to clone and mirror all data and code used for this inbox