public inbox for [email protected]
help / color / mirror / Atom feed[PATCH 3/3] Extended statistics on expressions
26+ messages / 7 participants
[nested] [flat]
* [PATCH 3/3] Extended statistics on expressions
@ 2020-12-03 15:19 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 26+ messages in thread
From: Tomas Vondra @ 2020-12-03 15:19 UTC (permalink / raw)
Allow defining extended statistics on expressions, not just simple
column references. With this commit, it's possible to do things like
CREATE TABLE t (a int);
CREATE STATISTICS s ON mod(a,10), mod(a,20) FROM t;
and the collected statistics will be useful for estimating queries
using those expressions in various places, like
SELECT * FROM t WHERE mod(a,10) = 0 AND mod(a,20) = 0;
or
SELECT mod(a,10), mod(a,20) FROM t GROUP BY 1, 2;
The commit also adds a new statistics type "expressions" which builds
the usual per-column statistics for each expression, allowing better
estimates even for queries with just a single expression, which are
not affected by multi-column statistics. This achieves the same goal
as creating expression indexes, without index maintenance overhead.
---
doc/src/sgml/catalogs.sgml | 238 ++-
doc/src/sgml/ref/create_statistics.sgml | 98 +-
src/backend/catalog/Makefile | 8 +-
src/backend/catalog/system_views.sql | 71 +
src/backend/commands/statscmds.c | 319 +++-
src/backend/nodes/copyfuncs.c | 14 +
src/backend/nodes/equalfuncs.c | 13 +
src/backend/nodes/outfuncs.c | 12 +
src/backend/optimizer/util/plancat.c | 62 +
src/backend/parser/gram.y | 38 +-
src/backend/parser/parse_agg.c | 10 +
src/backend/parser/parse_expr.c | 6 +
src/backend/parser/parse_func.c | 3 +
src/backend/parser/parse_utilcmd.c | 121 +-
src/backend/statistics/dependencies.c | 369 +++-
src/backend/statistics/extended_stats.c | 1558 ++++++++++++++++-
src/backend/statistics/mcv.c | 295 +++-
src/backend/statistics/mvdistinct.c | 101 +-
src/backend/tcop/utility.c | 23 +-
src/backend/utils/adt/ruleutils.c | 269 ++-
src/backend/utils/adt/selfuncs.c | 447 ++++-
src/bin/pg_dump/t/002_pg_dump.pl | 12 +
src/bin/psql/describe.c | 66 +-
src/include/catalog/pg_proc.dat | 8 +
src/include/catalog/pg_statistic_ext.h | 4 +
src/include/catalog/pg_statistic_ext_data.h | 1 +
src/include/nodes/nodes.h | 1 +
src/include/nodes/parsenodes.h | 16 +
src/include/nodes/pathnodes.h | 3 +-
src/include/parser/parse_node.h | 1 +
src/include/parser/parse_utilcmd.h | 2 +
.../statistics/extended_stats_internal.h | 40 +-
src/include/statistics/statistics.h | 2 +
.../regress/expected/create_table_like.out | 20 +-
src/test/regress/expected/rules.out | 75 +
src/test/regress/expected/stats_ext.out | 681 ++++++-
src/test/regress/sql/create_table_like.sql | 2 +
src/test/regress/sql/stats_ext.sql | 313 +++-
38 files changed, 4944 insertions(+), 378 deletions(-)
diff --git a/doc/src/sgml/catalogs.sgml b/doc/src/sgml/catalogs.sgml
index 43d7a1ad90..eef546a23f 100644
--- a/doc/src/sgml/catalogs.sgml
+++ b/doc/src/sgml/catalogs.sgml
@@ -7347,7 +7347,7 @@ SCRAM-SHA-256$<replaceable><iteration count></replaceable>:<replaceable>&l
<structfield>stxkind</structfield> <type>char[]</type>
</para>
<para>
- An array containing codes for the enabled statistic kinds;
+ An array containing codes for the enabled statistics kinds;
valid values are:
<literal>d</literal> for n-distinct statistics,
<literal>f</literal> for functional dependency statistics, and
@@ -9396,6 +9396,11 @@ SCRAM-SHA-256$<replaceable><iteration count></replaceable>:<replaceable>&l
<entry>extended planner statistics</entry>
</row>
+ <row>
+ <entry><link linkend="view-pg-stats-ext-exprs"><structname>pg_stats_ext_exprs</structname></link></entry>
+ <entry>extended planner statistics for expressions</entry>
+ </row>
+
<row>
<entry><link linkend="view-pg-tables"><structname>pg_tables</structname></link></entry>
<entry>tables</entry>
@@ -12958,6 +12963,237 @@ SELECT * FROM pg_locks pl LEFT JOIN pg_prepared_xacts ppx
</sect1>
+ <sect1 id="view-pg-stats-ext-exprs">
+ <title><structname>pg_stats_ext_exprs</structname></title>
+
+ <indexterm zone="view-pg-stats-ext-exprs">
+ <primary>pg_stats_ext_exprs</primary>
+ </indexterm>
+
+ <para>
+ The view <structname>pg_stats_ext_exprs</structname> provides access to
+ the information stored in the <link
+ linkend="catalog-pg-statistic-ext"><structname>pg_statistic_ext</structname></link>
+ and <link linkend="catalog-pg-statistic-ext-data"><structname>pg_statistic_ext_data</structname></link>
+ catalogs. This view allows access only to rows of
+ <link linkend="catalog-pg-statistic-ext"><structname>pg_statistic_ext</structname></link> and <link linkend="catalog-pg-statistic-ext-data"><structname>pg_statistic_ext_data</structname></link>
+ that correspond to tables the user has permission to read, and therefore
+ it is safe to allow public read access to this view.
+ </para>
+
+ <para>
+ <structname>pg_stats_ext_exprs</structname> is also designed to present
+ the information in a more readable format than the underlying catalogs
+ — at the cost that its schema must be extended whenever the structure
+ of statistics in <link linkend="catalog-pg-statistic"><structname>pg_statistic</structname></link> changes.
+ </para>
+
+ <table>
+ <title><structname>pg_stats_ext_exprs</structname> Columns</title>
+ <tgroup cols="1">
+ <thead>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ Column Type
+ </para>
+ <para>
+ Description
+ </para></entry>
+ </row>
+ </thead>
+
+ <tbody>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>schemaname</structfield> <type>name</type>
+ (references <link linkend="catalog-pg-namespace"><structname>pg_namespace</structname></link>.<structfield>nspname</structfield>)
+ </para>
+ <para>
+ Name of schema containing table
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>tablename</structfield> <type>name</type>
+ (references <link linkend="catalog-pg-class"><structname>pg_class</structname></link>.<structfield>relname</structfield>)
+ </para>
+ <para>
+ Name of table
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>statistics_schemaname</structfield> <type>name</type>
+ (references <link linkend="catalog-pg-namespace"><structname>pg_namespace</structname></link>.<structfield>nspname</structfield>)
+ </para>
+ <para>
+ Name of schema containing extended statistic
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>statistics_name</structfield> <type>name</type>
+ (references <link linkend="catalog-pg-statistic-ext"><structname>pg_statistic_ext</structname></link>.<structfield>stxname</structfield>)
+ </para>
+ <para>
+ Name of extended statistics
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>statistics_owner</structfield> <type>name</type>
+ (references <link linkend="catalog-pg-authid"><structname>pg_authid</structname></link>.<structfield>rolname</structfield>)
+ </para>
+ <para>
+ Owner of the extended statistics
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>expr</structfield> <type>text</type>
+ </para>
+ <para>
+ Expression the extended statistics is defined on
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>null_frac</structfield> <type>float4</type>
+ </para>
+ <para>
+ Fraction of column entries that are null
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>avg_width</structfield> <type>int4</type>
+ </para>
+ <para>
+ Average width in bytes of column's entries
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>n_distinct</structfield> <type>float4</type>
+ </para>
+ <para>
+ If greater than zero, the estimated number of distinct values in the
+ column. If less than zero, the negative of the number of distinct
+ values divided by the number of rows. (The negated form is used when
+ <command>ANALYZE</command> believes that the number of distinct values is
+ likely to increase as the table grows; the positive form is used when
+ the column seems to have a fixed number of possible values.) For
+ example, -1 indicates a unique column in which the number of distinct
+ values is the same as the number of rows.
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>most_common_vals</structfield> <type>anyarray</type>
+ </para>
+ <para>
+ A list of the most common values in the column. (Null if
+ no values seem to be more common than any others.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>most_common_freqs</structfield> <type>float4[]</type>
+ </para>
+ <para>
+ A list of the frequencies of the most common values,
+ i.e., number of occurrences of each divided by total number of rows.
+ (Null when <structfield>most_common_vals</structfield> is.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>histogram_bounds</structfield> <type>anyarray</type>
+ </para>
+ <para>
+ A list of values that divide the column's values into groups of
+ approximately equal population. The values in
+ <structfield>most_common_vals</structfield>, if present, are omitted from this
+ histogram calculation. (This column is null if the column data type
+ does not have a <literal><</literal> operator or if the
+ <structfield>most_common_vals</structfield> list accounts for the entire
+ population.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>correlation</structfield> <type>float4</type>
+ </para>
+ <para>
+ Statistical correlation between physical row ordering and
+ logical ordering of the column values. This ranges from -1 to +1.
+ When the value is near -1 or +1, an index scan on the column will
+ be estimated to be cheaper than when it is near zero, due to reduction
+ of random access to the disk. (This column is null if the column data
+ type does not have a <literal><</literal> operator.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>most_common_elems</structfield> <type>anyarray</type>
+ </para>
+ <para>
+ A list of non-null element values most often appearing within values of
+ the column. (Null for scalar types.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>most_common_elem_freqs</structfield> <type>float4[]</type>
+ </para>
+ <para>
+ A list of the frequencies of the most common element values, i.e., the
+ fraction of rows containing at least one instance of the given value.
+ Two or three additional values follow the per-element frequencies;
+ these are the minimum and maximum of the preceding per-element
+ frequencies, and optionally the frequency of null elements.
+ (Null when <structfield>most_common_elems</structfield> is.)
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>elem_count_histogram</structfield> <type>float4[]</type>
+ </para>
+ <para>
+ A histogram of the counts of distinct non-null element values within the
+ values of the column, followed by the average number of distinct
+ non-null elements. (Null for scalar types.)
+ </para></entry>
+ </row>
+ </tbody>
+ </tgroup>
+ </table>
+
+ <para>
+ The maximum number of entries in the array fields can be controlled on a
+ column-by-column basis using the <link linkend="sql-altertable"><command>ALTER
+ TABLE SET STATISTICS</command></link>
+ command, or globally by setting the
+ <xref linkend="guc-default-statistics-target"/> run-time parameter.
+ </para>
+
+ </sect1>
+
<sect1 id="view-pg-tables">
<title><structname>pg_tables</structname></title>
diff --git a/doc/src/sgml/ref/create_statistics.sgml b/doc/src/sgml/ref/create_statistics.sgml
index 4363be50c3..ba50ee6bcd 100644
--- a/doc/src/sgml/ref/create_statistics.sgml
+++ b/doc/src/sgml/ref/create_statistics.sgml
@@ -21,9 +21,13 @@ PostgreSQL documentation
<refsynopsisdiv>
<synopsis>
+CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_name</replaceable>
+ ON ( <replaceable class="parameter">expression</replaceable> )
+ FROM <replaceable class="parameter">table_name</replaceable>
+
CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_name</replaceable>
[ ( <replaceable class="parameter">statistics_kind</replaceable> [, ... ] ) ]
- ON <replaceable class="parameter">column_name</replaceable>, <replaceable class="parameter">column_name</replaceable> [, ...]
+ ON { <replaceable class="parameter">column_name</replaceable> | ( <replaceable class="parameter">expression</replaceable> ) } [, ...]
FROM <replaceable class="parameter">table_name</replaceable>
</synopsis>
@@ -39,6 +43,16 @@ CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_na
database and will be owned by the user issuing the command.
</para>
+ <para>
+ The <command>CREATE STATISTICS</command> command has two basic forms. The
+ simple variant allows building statistics for a single expression, does
+ not allow specifying any statistics kinds and provides benefits similar
+ to an expression index. The full variant allows defining statistics objects
+ on multiple columns and expressions, and selecting which statistics kinds will
+ be built. The per-expression statistics are built automatically when there
+ is at least one expression.
+ </para>
+
<para>
If a schema name is given (for example, <literal>CREATE STATISTICS
myschema.mystat ...</literal>) then the statistics object is created in the
@@ -86,7 +100,9 @@ CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_na
dependency statistics, and <literal>mcv</literal> which enables
most-common values lists.
If this clause is omitted, all supported statistics kinds are
- included in the statistics object.
+ included in the statistics object. Expression statistics are built
+ automatically when the statistics definition includes complex
+ expressions and not just simple column references.
For more information, see <xref linkend="planner-stats-extended"/>
and <xref linkend="multivariate-statistics-examples"/>.
</para>
@@ -104,6 +120,17 @@ CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_na
</listitem>
</varlistentry>
+ <varlistentry>
+ <term><replaceable class="parameter">expression</replaceable></term>
+ <listitem>
+ <para>
+ The expression to be covered by the computed statistics. In this case
+ only a single expression is required, in which case only statistics
+ for the expression are built.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry>
<term><replaceable class="parameter">table_name</replaceable></term>
<listitem>
@@ -125,6 +152,13 @@ CREATE STATISTICS [ IF NOT EXISTS ] <replaceable class="parameter">statistics_na
reading it. Once created, however, the ownership of the statistics
object is independent of the underlying table(s).
</para>
+
+ <para>
+ Expression statistics are per-expression and are similar to creating an
+ index on the expression, except that they avoid the overhead of index
+ maintenance. Expression statistics are built automatically when there
+ is at least one expression in the statistics object definition.
+ </para>
</refsect1>
<refsect1 id="sql-createstatistics-examples">
@@ -196,6 +230,66 @@ EXPLAIN ANALYZE SELECT * FROM t2 WHERE (a = 1) AND (b = 2);
in the table, allowing it to generate better estimates in both cases.
</para>
+ <para>
+ Create table <structname>t3</structname> with a single timestamp column,
+ and run a query using an expression on that column. Without extended
+ statistics, the planner has no information about data distribution for
+ results of those expression, and uses default estimates as illustrated
+ by the first query. The planner also does not realize that the value of
+ the second column fully defines the value of the other column, because
+ date truncated to day still identifies the month. Then expression and
+ ndistinct statistics are built on those two columns:
+
+<programlisting>
+CREATE TABLE t3 (
+ a timestamp
+);
+
+INSERT INTO t3 SELECT i FROM generate_series('2020-01-01'::timestamp,
+ '2020-12-31'::timestamp,
+ '1 minute'::interval) s(i);
+
+ANALYZE t3;
+
+-- the number of matching rows will be drastically underestimated:
+EXPLAIN ANALYZE SELECT * FROM t3
+ WHERE date_trunc('month', a) = '2020-01-01'::timestamp;
+
+EXPLAIN ANALYZE SELECT * FROM t3
+ WHERE date_trunc('day', a) BETWEEN '2020-01-01'::timestamp
+ AND '2020-06-30'::timestamp;
+
+EXPLAIN ANALYZE SELECT date_trunc('month', a), date_trunc('day', a)
+ FROM t3 GROUP BY 1, 2;
+
+-- per-expression statistics are built automatically
+CREATE STATISTICS s3 (ndistinct) ON date_trunc('month', a), date_trunc('day', a) FROM t3;
+
+ANALYZE t1;
+
+-- now the row count estimates are more accurate:
+EXPLAIN ANALYZE SELECT * FROM t3
+ WHERE date_trunc('month', a) = '2020-01-01'::timestamp;
+
+EXPLAIN ANALYZE SELECT * FROM t3
+ WHERE date_trunc('day', a) BETWEEN '2020-01-01'::timestamp
+ AND '2020-06-30'::timestamp;
+
+EXPLAIN ANALYZE SELECT date_trunc('month', a), date_trunc('day', a)
+ FROM t3 GROUP BY 1, 2;
+</programlisting>
+
+ Without expression and ndistinct statistics, the planner would assume
+ that the two <literal>WHERE</literal> and <literal>GROUP BY</literal>
+ conditions are independent, and would multiply their selectivities
+ together to arrive at a much-too-small row count estimate in the first
+ two queries, and a much-too-high group count estimate in the aggregate
+ query. This is further exacerbated by the lack of accurate statistics
+ for the expressions, forcing the planner to use default selectivities.
+ With such statistics, the planner recognizes that the conditions are
+ correlated and arrives at much more accurate estimates.
+ </para>
+
</refsect1>
<refsect1>
diff --git a/src/backend/catalog/Makefile b/src/backend/catalog/Makefile
index c85f0ca7b6..fa91ff1c42 100644
--- a/src/backend/catalog/Makefile
+++ b/src/backend/catalog/Makefile
@@ -49,15 +49,15 @@ include $(top_srcdir)/src/backend/common.mk
# Note: the order of this list determines the order in which the catalog
# header files are assembled into postgres.bki. BKI_BOOTSTRAP catalogs
-# must appear first, and there are reputedly other, undocumented ordering
-# dependencies.
+# must appear first, and pg_statistic before pg_statistic_ext_data, and
+# there are reputedly other, undocumented ordering dependencies.
CATALOG_HEADERS := \
pg_proc.h pg_type.h pg_attribute.h pg_class.h \
pg_attrdef.h pg_constraint.h pg_inherits.h pg_index.h pg_operator.h \
pg_opfamily.h pg_opclass.h pg_am.h pg_amop.h pg_amproc.h \
pg_language.h pg_largeobject_metadata.h pg_largeobject.h pg_aggregate.h \
- pg_statistic_ext.h pg_statistic_ext_data.h \
- pg_statistic.h pg_rewrite.h pg_trigger.h pg_event_trigger.h pg_description.h \
+ pg_statistic.h pg_statistic_ext.h pg_statistic_ext_data.h \
+ pg_rewrite.h pg_trigger.h pg_event_trigger.h pg_description.h \
pg_cast.h pg_enum.h pg_namespace.h pg_conversion.h pg_depend.h \
pg_database.h pg_db_role_setting.h pg_tablespace.h \
pg_authid.h pg_auth_members.h pg_shdepend.h pg_shdescription.h \
diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index fa58afd9d7..32ad93db3f 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -264,6 +264,7 @@ CREATE VIEW pg_stats_ext WITH (security_barrier) AS
JOIN pg_attribute a
ON (a.attrelid = s.stxrelid AND a.attnum = k)
) AS attnames,
+ pg_get_statisticsobjdef_expressions(s.oid) as exprs,
s.stxkind AS kinds,
sd.stxdndistinct AS n_distinct,
sd.stxddependencies AS dependencies,
@@ -290,6 +291,76 @@ CREATE VIEW pg_stats_ext WITH (security_barrier) AS
WHERE NOT has_column_privilege(c.oid, a.attnum, 'select') )
AND (c.relrowsecurity = false OR NOT row_security_active(c.oid));
+CREATE VIEW pg_stats_ext_exprs WITH (security_barrier) AS
+ SELECT cn.nspname AS schemaname,
+ c.relname AS tablename,
+ sn.nspname AS statistics_schemaname,
+ s.stxname AS statistics_name,
+ pg_get_userbyid(s.stxowner) AS statistics_owner,
+ stat_exprs.expr,
+ (stat.a).stanullfrac AS null_frac,
+ (stat.a).stawidth AS avg_width,
+ (stat.a).stadistinct AS n_distinct,
+ (CASE
+ WHEN (stat.a).stakind1 = 1 THEN (stat.a).stavalues1
+ WHEN (stat.a).stakind2 = 1 THEN (stat.a).stavalues2
+ WHEN (stat.a).stakind3 = 1 THEN (stat.a).stavalues3
+ WHEN (stat.a).stakind4 = 1 THEN (stat.a).stavalues4
+ WHEN (stat.a).stakind5 = 1 THEN (stat.a).stavalues5
+ END) AS most_common_vals,
+ (CASE
+ WHEN (stat.a).stakind1 = 1 THEN (stat.a).stanumbers1
+ WHEN (stat.a).stakind2 = 1 THEN (stat.a).stanumbers2
+ WHEN (stat.a).stakind3 = 1 THEN (stat.a).stanumbers3
+ WHEN (stat.a).stakind4 = 1 THEN (stat.a).stanumbers4
+ WHEN (stat.a).stakind5 = 1 THEN (stat.a).stanumbers5
+ END) AS most_common_freqs,
+ (CASE
+ WHEN (stat.a).stakind1 = 2 THEN (stat.a).stavalues1
+ WHEN (stat.a).stakind2 = 2 THEN (stat.a).stavalues2
+ WHEN (stat.a).stakind3 = 2 THEN (stat.a).stavalues3
+ WHEN (stat.a).stakind4 = 2 THEN (stat.a).stavalues4
+ WHEN (stat.a).stakind5 = 2 THEN (stat.a).stavalues5
+ END) AS histogram_bounds,
+ (CASE
+ WHEN (stat.a).stakind1 = 3 THEN (stat.a).stanumbers1[1]
+ WHEN (stat.a).stakind2 = 3 THEN (stat.a).stanumbers2[1]
+ WHEN (stat.a).stakind3 = 3 THEN (stat.a).stanumbers3[1]
+ WHEN (stat.a).stakind4 = 3 THEN (stat.a).stanumbers4[1]
+ WHEN (stat.a).stakind5 = 3 THEN (stat.a).stanumbers5[1]
+ END) correlation,
+ (CASE
+ WHEN (stat.a).stakind1 = 4 THEN (stat.a).stavalues1
+ WHEN (stat.a).stakind2 = 4 THEN (stat.a).stavalues2
+ WHEN (stat.a).stakind3 = 4 THEN (stat.a).stavalues3
+ WHEN (stat.a).stakind4 = 4 THEN (stat.a).stavalues4
+ WHEN (stat.a).stakind5 = 4 THEN (stat.a).stavalues5
+ END) AS most_common_elems,
+ (CASE
+ WHEN (stat.a).stakind1 = 4 THEN (stat.a).stanumbers1
+ WHEN (stat.a).stakind2 = 4 THEN (stat.a).stanumbers2
+ WHEN (stat.a).stakind3 = 4 THEN (stat.a).stanumbers3
+ WHEN (stat.a).stakind4 = 4 THEN (stat.a).stanumbers4
+ WHEN (stat.a).stakind5 = 4 THEN (stat.a).stanumbers5
+ END) AS most_common_elem_freqs,
+ (CASE
+ WHEN (stat.a).stakind1 = 5 THEN (stat.a).stanumbers1
+ WHEN (stat.a).stakind2 = 5 THEN (stat.a).stanumbers2
+ WHEN (stat.a).stakind3 = 5 THEN (stat.a).stanumbers3
+ WHEN (stat.a).stakind4 = 5 THEN (stat.a).stanumbers4
+ WHEN (stat.a).stakind5 = 5 THEN (stat.a).stanumbers5
+ END) AS elem_count_histogram
+ FROM pg_statistic_ext s JOIN pg_class c ON (c.oid = s.stxrelid)
+ LEFT JOIN pg_statistic_ext_data sd ON (s.oid = sd.stxoid)
+ LEFT JOIN pg_namespace cn ON (cn.oid = c.relnamespace)
+ LEFT JOIN pg_namespace sn ON (sn.oid = s.stxnamespace)
+ JOIN LATERAL (
+ SELECT unnest(pg_get_statisticsobjdef_expressions(s.oid)) AS expr
+ ) stat_exprs ON (stat_exprs.expr IS NOT NULL)
+ LEFT JOIN LATERAL (
+ SELECT unnest(sd.stxdexpr)::pg_statistic AS a
+ ) stat ON (TRUE);
+
-- unprivileged users may read pg_statistic_ext but not pg_statistic_ext_data
REVOKE ALL on pg_statistic_ext_data FROM public;
diff --git a/src/backend/commands/statscmds.c b/src/backend/commands/statscmds.c
index 2bae205845..7370af820f 100644
--- a/src/backend/commands/statscmds.c
+++ b/src/backend/commands/statscmds.c
@@ -29,6 +29,8 @@
#include "commands/comment.h"
#include "commands/defrem.h"
#include "miscadmin.h"
+#include "nodes/nodeFuncs.h"
+#include "optimizer/optimizer.h"
#include "statistics/statistics.h"
#include "utils/builtins.h"
#include "utils/fmgroids.h"
@@ -62,7 +64,8 @@ ObjectAddress
CreateStatistics(CreateStatsStmt *stmt)
{
int16 attnums[STATS_MAX_DIMENSIONS];
- int numcols = 0;
+ int nattnums = 0;
+ int numcols;
char *namestr;
NameData stxname;
Oid statoid;
@@ -74,21 +77,25 @@ CreateStatistics(CreateStatsStmt *stmt)
Datum datavalues[Natts_pg_statistic_ext_data];
bool datanulls[Natts_pg_statistic_ext_data];
int2vector *stxkeys;
+ List *stxexprs = NIL;
+ Datum exprsDatum;
Relation statrel;
Relation datarel;
Relation rel = NULL;
Oid relid;
ObjectAddress parentobject,
myself;
- Datum types[3]; /* one for each possible type of statistic */
+ Datum types[4]; /* one for each possible type of statistic */
int ntypes;
ArrayType *stxkind;
bool build_ndistinct;
bool build_dependencies;
bool build_mcv;
+ bool build_expressions;
bool requested_type = false;
int i;
ListCell *cell;
+ ListCell *cell2;
Assert(IsA(stmt, CreateStatsStmt));
@@ -190,72 +197,169 @@ CreateStatistics(CreateStatsStmt *stmt)
}
/*
- * Currently, we only allow simple column references in the expression
- * list. That will change someday, and again the grammar already supports
- * it so we have to enforce restrictions here. For now, we can convert
- * the expression list to a simple array of attnums. While at it, enforce
- * some constraints.
+ * Make sure no more than STATS_MAX_DIMENSIONS columns are used. There
+ * might be duplicates and so on, but we'll deal with those later.
+ */
+ numcols = list_length(stmt->exprs);
+ if (numcols > STATS_MAX_DIMENSIONS)
+ ereport(ERROR,
+ (errcode(ERRCODE_TOO_MANY_COLUMNS),
+ errmsg("cannot have more than %d columns in statistics",
+ STATS_MAX_DIMENSIONS)));
+
+ /*
+ * Convert the expression list to a simple array of attnums, but also
+ * keep a list of more complex expressions. While at it, enforce some
+ * constraints.
*/
foreach(cell, stmt->exprs)
{
Node *expr = (Node *) lfirst(cell);
- ColumnRef *cref;
- char *attname;
+ StatsElem *selem;
HeapTuple atttuple;
Form_pg_attribute attForm;
TypeCacheEntry *type;
- if (!IsA(expr, ColumnRef))
+ /*
+ * XXX How could we get anything else than a StatsElem, given the
+ * grammar? But let's keep it as a safety, maybe shall we turn it
+ * into an assert?
+ */
+ if (!IsA(expr, StatsElem))
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("only simple column references are allowed in CREATE STATISTICS")));
- cref = (ColumnRef *) expr;
+ errmsg("only simple column references and expressions are allowed in CREATE STATISTICS")));
- if (list_length(cref->fields) != 1)
- ereport(ERROR,
- (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("only simple column references are allowed in CREATE STATISTICS")));
- attname = strVal((Value *) linitial(cref->fields));
+ selem = (StatsElem *) expr;
- atttuple = SearchSysCacheAttName(relid, attname);
- if (!HeapTupleIsValid(atttuple))
- ereport(ERROR,
- (errcode(ERRCODE_UNDEFINED_COLUMN),
- errmsg("column \"%s\" does not exist",
- attname)));
- attForm = (Form_pg_attribute) GETSTRUCT(atttuple);
+ if (selem->name) /* column reference */
+ {
+ char *attname;
+ attname = selem->name;
+
+ atttuple = SearchSysCacheAttName(relid, attname);
+ if (!HeapTupleIsValid(atttuple))
+ ereport(ERROR,
+ (errcode(ERRCODE_UNDEFINED_COLUMN),
+ errmsg("column \"%s\" does not exist",
+ attname)));
+ attForm = (Form_pg_attribute) GETSTRUCT(atttuple);
+
+ /* Disallow use of system attributes in extended stats */
+ if (attForm->attnum <= 0)
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("statistics creation on system columns is not supported")));
+
+ /* Disallow data types without a less-than operator */
+ type = lookup_type_cache(attForm->atttypid, TYPECACHE_LT_OPR);
+ if (type->lt_opr == InvalidOid)
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("column \"%s\" cannot be used in statistics because its type %s has no default btree operator class",
+ attname, format_type_be(attForm->atttypid))));
+
+ attnums[nattnums] = attForm->attnum;
+ nattnums++;
+ ReleaseSysCache(atttuple);
+ }
+ else /* expression */
+ {
+ Node *expr = selem->expr;
+ Oid atttype;
+
+ Assert(expr != NULL);
+
+ /*
+ * Disallow data types without a less-than operator.
+ *
+ * We ignore this for statistics on a single expression, in
+ * which case we'll build the regular statistics only (and
+ * that code can deal with such data types).
+ */
+ if (list_length(stmt->exprs) > 1)
+ {
+ atttype = exprType(expr);
+ type = lookup_type_cache(atttype, TYPECACHE_LT_OPR);
+ if (type->lt_opr == InvalidOid)
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("expression cannot be used in statistics because its type %s has no default btree operator class",
+ format_type_be(atttype))));
+ }
+
+ stxexprs = lappend(stxexprs, expr);
+ }
+ }
- /* Disallow use of system attributes in extended stats */
- if (attForm->attnum <= 0)
+ /*
+ * Parse the statistics kinds. Firstly, check that this is not the
+ * variant building statistics for a single expression, in which case
+ * we don't allow specifying any statistics kinds. The simple variant
+ * only has one expression, and does not allow statistics kinds.
+ */
+ if ((list_length(stmt->exprs) == 1) && (list_length(stxexprs) == 1))
+ {
+ /* statistics kinds not specified */
+ if (list_length(stmt->stat_types) > 0)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("statistics creation on system columns is not supported")));
+ errmsg("when building statistics on a single expression, statistics kinds may not be specified")));
+ }
- /* Disallow data types without a less-than operator */
- type = lookup_type_cache(attForm->atttypid, TYPECACHE_LT_OPR);
- if (type->lt_opr == InvalidOid)
- ereport(ERROR,
- (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("column \"%s\" cannot be used in statistics because its type %s has no default btree operator class",
- attname, format_type_be(attForm->atttypid))));
+ /* OK, let's check that we recognize the statistics kinds. */
+ build_ndistinct = false;
+ build_dependencies = false;
+ build_mcv = false;
+ foreach(cell, stmt->stat_types)
+ {
+ char *type = strVal((Value *) lfirst(cell));
- /* Make sure no more than STATS_MAX_DIMENSIONS columns are used */
- if (numcols >= STATS_MAX_DIMENSIONS)
+ if (strcmp(type, "ndistinct") == 0)
+ {
+ build_ndistinct = true;
+ requested_type = true;
+ }
+ else if (strcmp(type, "dependencies") == 0)
+ {
+ build_dependencies = true;
+ requested_type = true;
+ }
+ else if (strcmp(type, "mcv") == 0)
+ {
+ build_mcv = true;
+ requested_type = true;
+ }
+ else
ereport(ERROR,
- (errcode(ERRCODE_TOO_MANY_COLUMNS),
- errmsg("cannot have more than %d columns in statistics",
- STATS_MAX_DIMENSIONS)));
+ (errcode(ERRCODE_SYNTAX_ERROR),
+ errmsg("unrecognized statistics kind \"%s\"",
+ type)));
+ }
- attnums[numcols] = attForm->attnum;
- numcols++;
- ReleaseSysCache(atttuple);
+ /*
+ * If no statistic type was specified, build them all (but request
+ * expression stats only when there actually are any expressions).
+ */
+ if (!requested_type)
+ {
+ build_ndistinct = (numcols >= 2);
+ build_dependencies = (numcols >= 2);
+ build_mcv = (numcols >= 2);
}
/*
- * Check that at least two columns were specified in the statement. The
- * upper bound was already checked in the loop above.
+ * When there are non-trivial expressions, build the expression stats
+ * automatically. This allows calculating good estimates for stats that
+ * consider per-clause estimates (e.g. functional dependencies).
+ */
+ build_expressions = (list_length(stxexprs) > 0);
+
+ /*
+ * Check that at least two columns were specified in the statement, or
+ * that we're building statistics on a single expression.
*/
- if (numcols < 2)
+ if ((numcols < 2) && (list_length(stxexprs) != 1))
ereport(ERROR,
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
errmsg("extended statistics require at least 2 columns")));
@@ -265,13 +369,13 @@ CreateStatistics(CreateStatsStmt *stmt)
* it does not hurt (it does not affect the efficiency, unlike for
* indexes, for example).
*/
- qsort(attnums, numcols, sizeof(int16), compare_int16);
+ qsort(attnums, nattnums, sizeof(int16), compare_int16);
/*
* Check for duplicates in the list of columns. The attnums are sorted so
* just check consecutive elements.
*/
- for (i = 1; i < numcols; i++)
+ for (i = 1; i < nattnums; i++)
{
if (attnums[i] == attnums[i - 1])
ereport(ERROR,
@@ -279,48 +383,36 @@ CreateStatistics(CreateStatsStmt *stmt)
errmsg("duplicate column name in statistics definition")));
}
- /* Form an int2vector representation of the sorted column list */
- stxkeys = buildint2vector(attnums, numcols);
-
/*
- * Parse the statistics kinds.
+ * Check for duplicate expressions. We do two loops, counting the
+ * occurrences of each expression. This is O(N^2) but we only allow
+ * small number of expressions and it's not executed often.
*/
- build_ndistinct = false;
- build_dependencies = false;
- build_mcv = false;
- foreach(cell, stmt->stat_types)
+ foreach (cell, stxexprs)
{
- char *type = strVal((Value *) lfirst(cell));
+ Node *expr1 = (Node *) lfirst(cell);
+ int cnt = 0;
- if (strcmp(type, "ndistinct") == 0)
- {
- build_ndistinct = true;
- requested_type = true;
- }
- else if (strcmp(type, "dependencies") == 0)
+ foreach (cell2, stxexprs)
{
- build_dependencies = true;
- requested_type = true;
- }
- else if (strcmp(type, "mcv") == 0)
- {
- build_mcv = true;
- requested_type = true;
+ Node *expr2 = (Node *) lfirst(cell2);
+
+ if (equal(expr1, expr2))
+ cnt += 1;
}
- else
+
+ /* every expression should find at least itself */
+ Assert(cnt >= 1);
+
+ if (cnt > 1)
ereport(ERROR,
- (errcode(ERRCODE_SYNTAX_ERROR),
- errmsg("unrecognized statistics kind \"%s\"",
- type)));
- }
- /* If no statistic type was specified, build them all. */
- if (!requested_type)
- {
- build_ndistinct = true;
- build_dependencies = true;
- build_mcv = true;
+ (errcode(ERRCODE_DUPLICATE_COLUMN),
+ errmsg("duplicate expression in statistics definition")));
}
+ /* Form an int2vector representation of the sorted column list */
+ stxkeys = buildint2vector(attnums, nattnums);
+
/* construct the char array of enabled statistic types */
ntypes = 0;
if (build_ndistinct)
@@ -329,9 +421,23 @@ CreateStatistics(CreateStatsStmt *stmt)
types[ntypes++] = CharGetDatum(STATS_EXT_DEPENDENCIES);
if (build_mcv)
types[ntypes++] = CharGetDatum(STATS_EXT_MCV);
+ if (build_expressions)
+ types[ntypes++] = CharGetDatum(STATS_EXT_EXPRESSIONS);
Assert(ntypes > 0 && ntypes <= lengthof(types));
stxkind = construct_array(types, ntypes, CHAROID, 1, true, TYPALIGN_CHAR);
+ /* convert the expressions (if any) to a text datum */
+ if (stxexprs != NIL)
+ {
+ char *exprsString;
+
+ exprsString = nodeToString(stxexprs);
+ exprsDatum = CStringGetTextDatum(exprsString);
+ pfree(exprsString);
+ }
+ else
+ exprsDatum = (Datum) 0;
+
statrel = table_open(StatisticExtRelationId, RowExclusiveLock);
/*
@@ -351,6 +457,10 @@ CreateStatistics(CreateStatsStmt *stmt)
values[Anum_pg_statistic_ext_stxkeys - 1] = PointerGetDatum(stxkeys);
values[Anum_pg_statistic_ext_stxkind - 1] = PointerGetDatum(stxkind);
+ values[Anum_pg_statistic_ext_stxexprs - 1] = exprsDatum;
+ if (exprsDatum == (Datum) 0)
+ nulls[Anum_pg_statistic_ext_stxexprs - 1] = true;
+
/* insert it into pg_statistic_ext */
htup = heap_form_tuple(statrel->rd_att, values, nulls);
CatalogTupleInsert(statrel, htup);
@@ -373,6 +483,7 @@ CreateStatistics(CreateStatsStmt *stmt)
datanulls[Anum_pg_statistic_ext_data_stxdndistinct - 1] = true;
datanulls[Anum_pg_statistic_ext_data_stxddependencies - 1] = true;
datanulls[Anum_pg_statistic_ext_data_stxdmcv - 1] = true;
+ datanulls[Anum_pg_statistic_ext_data_stxdexpr - 1] = true;
/* insert it into pg_statistic_ext_data */
htup = heap_form_tuple(datarel->rd_att, datavalues, datanulls);
@@ -396,12 +507,39 @@ CreateStatistics(CreateStatsStmt *stmt)
*/
ObjectAddressSet(myself, StatisticExtRelationId, statoid);
- for (i = 0; i < numcols; i++)
+ /* add dependencies for plain column references */
+ for (i = 0; i < nattnums; i++)
{
ObjectAddressSubSet(parentobject, RelationRelationId, relid, attnums[i]);
recordDependencyOn(&myself, &parentobject, DEPENDENCY_AUTO);
}
+ /*
+ * If there are no simply-referenced columns, give the statistics an
+ * auto dependency on the whole table. In most cases, this will
+ * be redundant, but it might not be if the statistics expressions
+ * contain no Vars (which might seem strange but possible).
+ *
+ * XXX This is copied from index_create, not sure if it's applicable
+ * to extended statistics too.
+ */
+ if (!nattnums)
+ {
+ ObjectAddressSet(parentobject, RelationRelationId, relid);
+ recordDependencyOn(&myself, &parentobject, DEPENDENCY_AUTO);
+ }
+
+ /*
+ * Store dependencies on anything mentioned in statistics expressions,
+ * just like we do for index expressions.
+ */
+ if (stxexprs)
+ recordDependencyOnSingleRelExpr(&myself,
+ (Node *) stxexprs,
+ relid,
+ DEPENDENCY_NORMAL,
+ DEPENDENCY_AUTO, false, true);
+
/*
* Also add dependencies on namespace and owner. These are required
* because the stats object might have a different namespace and/or owner
@@ -625,7 +763,7 @@ UpdateStatisticsForTypeChange(Oid statsOid, Oid relationOid, int attnum,
elog(ERROR, "cache lookup failed for statistics object %u", statsOid);
/*
- * When none of the defined statistics types contain datum values from the
+ * When none of the defined statistics kinds contain datum values from the
* table's columns then there's no need to reset the stats. Functional
* dependencies and ndistinct stats should still hold true.
*/
@@ -637,7 +775,7 @@ UpdateStatisticsForTypeChange(Oid statsOid, Oid relationOid, int attnum,
/*
* OK, we need to reset some statistics. So let's build the new tuple,
- * replacing the affected statistics types with NULL.
+ * replacing the affected statistics kinds with NULL.
*/
memset(nulls, 0, Natts_pg_statistic_ext_data * sizeof(bool));
memset(replaces, 0, Natts_pg_statistic_ext_data * sizeof(bool));
@@ -645,6 +783,7 @@ UpdateStatisticsForTypeChange(Oid statsOid, Oid relationOid, int attnum,
replaces[Anum_pg_statistic_ext_data_stxdmcv - 1] = true;
nulls[Anum_pg_statistic_ext_data_stxdmcv - 1] = true;
+ nulls[Anum_pg_statistic_ext_data_stxdexpr - 1] = true;
rel = table_open(StatisticExtDataRelationId, RowExclusiveLock);
@@ -731,18 +870,26 @@ ChooseExtendedStatisticNameAddition(List *exprs)
buf[0] = '\0';
foreach(lc, exprs)
{
- ColumnRef *cref = (ColumnRef *) lfirst(lc);
+ StatsElem *selem = (StatsElem *) lfirst(lc);
const char *name;
/* It should be one of these, but just skip if it happens not to be */
- if (!IsA(cref, ColumnRef))
+ if (!IsA(selem, StatsElem))
continue;
- name = strVal((Value *) linitial(cref->fields));
+ name = selem->name;
if (buflen > 0)
buf[buflen++] = '_'; /* insert _ between names */
+ /*
+ * FIXME use 'expr' for expressions, which have empty column names.
+ * For indexes this is handled in ChooseIndexColumnNames, but we
+ * have no such function for stats.
+ */
+ if (!name)
+ name = "expr";
+
/*
* At this point we have buflen <= NAMEDATALEN. name should be less
* than NAMEDATALEN already, but use strlcpy for paranoia.
diff --git a/src/backend/nodes/copyfuncs.c b/src/backend/nodes/copyfuncs.c
index ba3ccc712c..a21be7ffb1 100644
--- a/src/backend/nodes/copyfuncs.c
+++ b/src/backend/nodes/copyfuncs.c
@@ -2925,6 +2925,17 @@ _copyIndexElem(const IndexElem *from)
return newnode;
}
+static StatsElem *
+_copyStatsElem(const StatsElem *from)
+{
+ StatsElem *newnode = makeNode(StatsElem);
+
+ COPY_STRING_FIELD(name);
+ COPY_NODE_FIELD(expr);
+
+ return newnode;
+}
+
static ColumnDef *
_copyColumnDef(const ColumnDef *from)
{
@@ -5636,6 +5647,9 @@ copyObjectImpl(const void *from)
case T_IndexElem:
retval = _copyIndexElem(from);
break;
+ case T_StatsElem:
+ retval = _copyStatsElem(from);
+ break;
case T_ColumnDef:
retval = _copyColumnDef(from);
break;
diff --git a/src/backend/nodes/equalfuncs.c b/src/backend/nodes/equalfuncs.c
index a2ef853dc2..2a5421c10f 100644
--- a/src/backend/nodes/equalfuncs.c
+++ b/src/backend/nodes/equalfuncs.c
@@ -2593,6 +2593,16 @@ _equalIndexElem(const IndexElem *a, const IndexElem *b)
return true;
}
+
+static bool
+_equalStatsElem(const StatsElem *a, const StatsElem *b)
+{
+ COMPARE_STRING_FIELD(name);
+ COMPARE_NODE_FIELD(expr);
+
+ return true;
+}
+
static bool
_equalColumnDef(const ColumnDef *a, const ColumnDef *b)
{
@@ -3689,6 +3699,9 @@ equal(const void *a, const void *b)
case T_IndexElem:
retval = _equalIndexElem(a, b);
break;
+ case T_StatsElem:
+ retval = _equalStatsElem(a, b);
+ break;
case T_ColumnDef:
retval = _equalColumnDef(a, b);
break;
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 8392be6d44..956e8d8151 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -2932,6 +2932,15 @@ _outIndexElem(StringInfo str, const IndexElem *node)
WRITE_ENUM_FIELD(nulls_ordering, SortByNulls);
}
+static void
+_outStatsElem(StringInfo str, const StatsElem *node)
+{
+ WRITE_NODE_TYPE("STATSELEM");
+
+ WRITE_STRING_FIELD(name);
+ WRITE_NODE_FIELD(expr);
+}
+
static void
_outQuery(StringInfo str, const Query *node)
{
@@ -4241,6 +4250,9 @@ outNode(StringInfo str, const void *obj)
case T_IndexElem:
_outIndexElem(str, obj);
break;
+ case T_StatsElem:
+ _outStatsElem(str, obj);
+ break;
case T_Query:
_outQuery(str, obj);
break;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index da322b453e..1e64d52c83 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -34,6 +34,7 @@
#include "foreign/fdwapi.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
+#include "nodes/nodeFuncs.h"
#include "nodes/supportnodes.h"
#include "optimizer/clauses.h"
#include "optimizer/cost.h"
@@ -1302,6 +1303,7 @@ get_relation_constraints(PlannerInfo *root,
static List *
get_relation_statistics(RelOptInfo *rel, Relation relation)
{
+ Index varno = rel->relid;
List *statoidlist;
List *stainfos = NIL;
ListCell *l;
@@ -1316,6 +1318,7 @@ get_relation_statistics(RelOptInfo *rel, Relation relation)
HeapTuple dtup;
Bitmapset *keys = NULL;
int i;
+ List *exprs = NIL;
htup = SearchSysCache1(STATEXTOID, ObjectIdGetDatum(statOid));
if (!HeapTupleIsValid(htup))
@@ -1334,6 +1337,49 @@ get_relation_statistics(RelOptInfo *rel, Relation relation)
for (i = 0; i < staForm->stxkeys.dim1; i++)
keys = bms_add_member(keys, staForm->stxkeys.values[i]);
+ /*
+ * preprocess expression (if any)
+ *
+ * FIXME Should we cache the result somewhere?
+ */
+ {
+ bool isnull;
+ Datum datum;
+
+ /* decode expression (if any) */
+ datum = SysCacheGetAttr(STATEXTOID, htup,
+ Anum_pg_statistic_ext_stxexprs, &isnull);
+
+ if (!isnull)
+ {
+ char *exprsString;
+
+ exprsString = TextDatumGetCString(datum);
+ exprs = (List *) stringToNode(exprsString);
+ pfree(exprsString);
+
+ /*
+ * Run the expressions through eval_const_expressions. This is not just an
+ * optimization, but is necessary, because the planner will be comparing
+ * them to similarly-processed qual clauses, and may fail to detect valid
+ * matches without this. We must not use canonicalize_qual, however,
+ * since these aren't qual expressions.
+ */
+ exprs = (List *) eval_const_expressions(NULL, (Node *) exprs);
+
+ /* May as well fix opfuncids too */
+ fix_opfuncids((Node *) exprs);
+
+ /*
+ * Modify the copies we obtain from the relcache to have the
+ * correct varno for the parent relation, so that they match up
+ * correctly against qual clauses.
+ */
+ if (varno != 1)
+ ChangeVarNodes((Node *) exprs, 1, varno, 0);
+ }
+ }
+
/* add one StatisticExtInfo for each kind built */
if (statext_is_kind_built(dtup, STATS_EXT_NDISTINCT))
{
@@ -1343,6 +1389,7 @@ get_relation_statistics(RelOptInfo *rel, Relation relation)
info->rel = rel;
info->kind = STATS_EXT_NDISTINCT;
info->keys = bms_copy(keys);
+ info->exprs = exprs;
stainfos = lappend(stainfos, info);
}
@@ -1355,6 +1402,7 @@ get_relation_statistics(RelOptInfo *rel, Relation relation)
info->rel = rel;
info->kind = STATS_EXT_DEPENDENCIES;
info->keys = bms_copy(keys);
+ info->exprs = exprs;
stainfos = lappend(stainfos, info);
}
@@ -1367,6 +1415,20 @@ get_relation_statistics(RelOptInfo *rel, Relation relation)
info->rel = rel;
info->kind = STATS_EXT_MCV;
info->keys = bms_copy(keys);
+ info->exprs = exprs;
+
+ stainfos = lappend(stainfos, info);
+ }
+
+ if (statext_is_kind_built(dtup, STATS_EXT_EXPRESSIONS))
+ {
+ StatisticExtInfo *info = makeNode(StatisticExtInfo);
+
+ info->statOid = statOid;
+ info->rel = rel;
+ info->kind = STATS_EXT_EXPRESSIONS;
+ info->keys = bms_copy(keys);
+ info->exprs = exprs;
stainfos = lappend(stainfos, info);
}
diff --git a/src/backend/parser/gram.y b/src/backend/parser/gram.y
index 31c95443a5..d219976b53 100644
--- a/src/backend/parser/gram.y
+++ b/src/backend/parser/gram.y
@@ -232,6 +232,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query);
WindowDef *windef;
JoinExpr *jexpr;
IndexElem *ielem;
+ StatsElem *selem;
Alias *alias;
RangeVar *range;
IntoClause *into;
@@ -396,7 +397,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query);
old_aggr_definition old_aggr_list
oper_argtypes RuleActionList RuleActionMulti
opt_column_list columnList opt_name_list
- sort_clause opt_sort_clause sortby_list index_params
+ sort_clause opt_sort_clause sortby_list index_params stats_params
opt_include opt_c_include index_including_params
name_list role_list from_clause from_list opt_array_bounds
qualified_name_list any_name any_name_list type_name_list
@@ -500,6 +501,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query);
%type <list> func_alias_clause
%type <sortby> sortby
%type <ielem> index_elem index_elem_options
+%type <selem> stats_param
%type <node> table_ref
%type <jexpr> joined_table
%type <range> relation_expr
@@ -4049,7 +4051,7 @@ ExistingIndex: USING INDEX name { $$ = $3; }
CreateStatsStmt:
CREATE STATISTICS any_name
- opt_name_list ON expr_list FROM from_list
+ opt_name_list ON stats_params FROM from_list
{
CreateStatsStmt *n = makeNode(CreateStatsStmt);
n->defnames = $3;
@@ -4061,7 +4063,7 @@ CreateStatsStmt:
$$ = (Node *)n;
}
| CREATE STATISTICS IF_P NOT EXISTS any_name
- opt_name_list ON expr_list FROM from_list
+ opt_name_list ON stats_params FROM from_list
{
CreateStatsStmt *n = makeNode(CreateStatsStmt);
n->defnames = $6;
@@ -4074,6 +4076,36 @@ CreateStatsStmt:
}
;
+/*
+ * Statistics attributes can be either simple column references, or arbitrary
+ * expressions in parens. For compatibility with index attributes permitted
+ * in CREATE INDEX, we allow an expression that's just a function call to be
+ * written without parens.
+ */
+
+stats_params: stats_param { $$ = list_make1($1); }
+ | stats_params ',' stats_param { $$ = lappend($1, $3); }
+ ;
+
+stats_param: ColId
+ {
+ $$ = makeNode(StatsElem);
+ $$->name = $1;
+ $$->expr = NULL;
+ }
+ | func_expr_windowless
+ {
+ $$ = makeNode(StatsElem);
+ $$->name = NULL;
+ $$->expr = $1;
+ }
+ | '(' a_expr ')'
+ {
+ $$ = makeNode(StatsElem);
+ $$->name = NULL;
+ $$->expr = $2;
+ }
+ ;
/*****************************************************************************
*
diff --git a/src/backend/parser/parse_agg.c b/src/backend/parser/parse_agg.c
index 588f005dd9..0b0841afb9 100644
--- a/src/backend/parser/parse_agg.c
+++ b/src/backend/parser/parse_agg.c
@@ -484,6 +484,13 @@ check_agglevels_and_constraints(ParseState *pstate, Node *expr)
else
err = _("grouping operations are not allowed in index predicates");
+ break;
+ case EXPR_KIND_STATS_EXPRESSION:
+ if (isAgg)
+ err = _("aggregate functions are not allowed in statistics expressions");
+ else
+ err = _("grouping operations are not allowed in statistics expressions");
+
break;
case EXPR_KIND_ALTER_COL_TRANSFORM:
if (isAgg)
@@ -906,6 +913,9 @@ transformWindowFuncCall(ParseState *pstate, WindowFunc *wfunc,
case EXPR_KIND_INDEX_EXPRESSION:
err = _("window functions are not allowed in index expressions");
break;
+ case EXPR_KIND_STATS_EXPRESSION:
+ err = _("window functions are not allowed in statistics expressions");
+ break;
case EXPR_KIND_INDEX_PREDICATE:
err = _("window functions are not allowed in index predicates");
break;
diff --git a/src/backend/parser/parse_expr.c b/src/backend/parser/parse_expr.c
index 379355f9bf..fcc1bb33d1 100644
--- a/src/backend/parser/parse_expr.c
+++ b/src/backend/parser/parse_expr.c
@@ -500,6 +500,7 @@ transformColumnRef(ParseState *pstate, ColumnRef *cref)
case EXPR_KIND_FUNCTION_DEFAULT:
case EXPR_KIND_INDEX_EXPRESSION:
case EXPR_KIND_INDEX_PREDICATE:
+ case EXPR_KIND_STATS_EXPRESSION:
case EXPR_KIND_ALTER_COL_TRANSFORM:
case EXPR_KIND_EXECUTE_PARAMETER:
case EXPR_KIND_TRIGGER_WHEN:
@@ -1739,6 +1740,9 @@ transformSubLink(ParseState *pstate, SubLink *sublink)
case EXPR_KIND_INDEX_PREDICATE:
err = _("cannot use subquery in index predicate");
break;
+ case EXPR_KIND_STATS_EXPRESSION:
+ err = _("cannot use subquery in statistics expression");
+ break;
case EXPR_KIND_ALTER_COL_TRANSFORM:
err = _("cannot use subquery in transform expression");
break;
@@ -3028,6 +3032,8 @@ ParseExprKindName(ParseExprKind exprKind)
return "index expression";
case EXPR_KIND_INDEX_PREDICATE:
return "index predicate";
+ case EXPR_KIND_STATS_EXPRESSION:
+ return "statistics expression";
case EXPR_KIND_ALTER_COL_TRANSFORM:
return "USING";
case EXPR_KIND_EXECUTE_PARAMETER:
diff --git a/src/backend/parser/parse_func.c b/src/backend/parser/parse_func.c
index 07d0013e84..652930ddf9 100644
--- a/src/backend/parser/parse_func.c
+++ b/src/backend/parser/parse_func.c
@@ -2503,6 +2503,9 @@ check_srf_call_placement(ParseState *pstate, Node *last_srf, int location)
case EXPR_KIND_INDEX_PREDICATE:
err = _("set-returning functions are not allowed in index predicates");
break;
+ case EXPR_KIND_STATS_EXPRESSION:
+ err = _("set-returning functions are not allowed in statistics expressions");
+ break;
case EXPR_KIND_ALTER_COL_TRANSFORM:
err = _("set-returning functions are not allowed in transform expressions");
break;
diff --git a/src/backend/parser/parse_utilcmd.c b/src/backend/parser/parse_utilcmd.c
index b31f3afa03..0028240d1a 100644
--- a/src/backend/parser/parse_utilcmd.c
+++ b/src/backend/parser/parse_utilcmd.c
@@ -1898,6 +1898,9 @@ generateClonedExtStatsStmt(RangeVar *heapRel, Oid heapRelid,
stat_types = lappend(stat_types, makeString("dependencies"));
else if (enabled[i] == STATS_EXT_MCV)
stat_types = lappend(stat_types, makeString("mcv"));
+ else if (enabled[i] == STATS_EXT_EXPRESSIONS)
+ /* expression stats are not exposed to users */
+ continue;
else
elog(ERROR, "unrecognized statistics kind %c", enabled[i]);
}
@@ -1905,14 +1908,43 @@ generateClonedExtStatsStmt(RangeVar *heapRel, Oid heapRelid,
/* Determine which columns the statistics are on */
for (i = 0; i < statsrec->stxkeys.dim1; i++)
{
- ColumnRef *cref = makeNode(ColumnRef);
+ StatsElem *selem = makeNode(StatsElem);
AttrNumber attnum = statsrec->stxkeys.values[i];
- cref->fields = list_make1(makeString(get_attname(heapRelid,
- attnum, false)));
- cref->location = -1;
+ selem->name = get_attname(heapRelid, attnum, false);
+ selem->expr = NULL;
- def_names = lappend(def_names, cref);
+ def_names = lappend(def_names, selem);
+ }
+
+ /*
+ * Now handle expressions, if there are any. The order does not
+ * matter for extended stats, so we simply append them after
+ * simple column references.
+ */
+ datum = SysCacheGetAttr(STATEXTOID, ht_stats,
+ Anum_pg_statistic_ext_stxexprs, &isnull);
+
+ if (!isnull)
+ {
+ ListCell *lc;
+ List *exprs = NIL;
+ char *exprsString;
+
+ exprsString = TextDatumGetCString(datum);
+ exprs = (List *) stringToNode(exprsString);
+
+ foreach(lc, exprs)
+ {
+ StatsElem *selem = makeNode(StatsElem);
+
+ selem->name = NULL;
+ selem->expr = (Node *) lfirst(lc);
+
+ def_names = lappend(def_names, selem);
+ }
+
+ pfree(exprsString);
}
/* finally, build the output node */
@@ -1923,6 +1955,7 @@ generateClonedExtStatsStmt(RangeVar *heapRel, Oid heapRelid,
stats->relations = list_make1(heapRel);
stats->stxcomment = NULL;
stats->if_not_exists = false;
+ stats->transformed = true; /* don't need transformStatsStmt */
/* Clean up */
ReleaseSysCache(ht_stats);
@@ -2847,6 +2880,84 @@ transformIndexStmt(Oid relid, IndexStmt *stmt, const char *queryString)
return stmt;
}
+/*
+ * transformStatsStmt - parse analysis for CREATE STATISTICS
+ *
+ * To avoid race conditions, it's important that this function rely only on
+ * the passed-in relid (and not on stmt->relation) to determine the target
+ * relation.
+ */
+CreateStatsStmt *
+transformStatsStmt(Oid relid, CreateStatsStmt *stmt, const char *queryString)
+{
+ ParseState *pstate;
+ ParseNamespaceItem *nsitem;
+ ListCell *l;
+ Relation rel;
+
+ /* Nothing to do if statement already transformed. */
+ if (stmt->transformed)
+ return stmt;
+
+ /*
+ * We must not scribble on the passed-in CreateStatsStmt, so copy it. (This is
+ * overkill, but easy.)
+ */
+ stmt = copyObject(stmt);
+
+ /* Set up pstate */
+ pstate = make_parsestate(NULL);
+ pstate->p_sourcetext = queryString;
+
+ /*
+ * Put the parent table into the rtable so that the expressions can refer
+ * to its fields without qualification. Caller is responsible for locking
+ * relation, but we still need to open it.
+ */
+ rel = relation_open(relid, NoLock);
+ nsitem = addRangeTableEntryForRelation(pstate, rel,
+ AccessShareLock,
+ NULL, false, true);
+
+ /* no to join list, yes to namespaces */
+ addNSItemToQuery(pstate, nsitem, false, true, true);
+
+ /* take care of any expressions */
+ foreach(l, stmt->exprs)
+ {
+ StatsElem *selem = (StatsElem *) lfirst(l);
+
+ if (selem->expr)
+ {
+ /* Now do parse transformation of the expression */
+ selem->expr = transformExpr(pstate, selem->expr,
+ EXPR_KIND_STATS_EXPRESSION);
+
+ /* We have to fix its collations too */
+ assign_expr_collations(pstate, selem->expr);
+ }
+ }
+
+ /*
+ * Check that only the base rel is mentioned. (This should be dead code
+ * now that add_missing_from is history.)
+ */
+ if (list_length(pstate->p_rtable) != 1)
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
+ errmsg("statistics expressions and predicates can refer only to the table being indexed")));
+
+ free_parsestate(pstate);
+
+ /* Close relation */
+ table_close(rel, NoLock);
+
+ /* Mark statement as successfully transformed */
+ stmt->transformed = true;
+
+ return stmt;
+}
+
/*
* transformRuleStmt -
diff --git a/src/backend/statistics/dependencies.c b/src/backend/statistics/dependencies.c
index f6e399b192..6bf3127bcc 100644
--- a/src/backend/statistics/dependencies.c
+++ b/src/backend/statistics/dependencies.c
@@ -70,15 +70,18 @@ static void generate_dependencies(DependencyGenerator state);
static DependencyGenerator DependencyGenerator_init(int n, int k);
static void DependencyGenerator_free(DependencyGenerator state);
static AttrNumber *DependencyGenerator_next(DependencyGenerator state);
-static double dependency_degree(int numrows, HeapTuple *rows, int k,
- AttrNumber *dependency, VacAttrStats **stats, Bitmapset *attrs);
+static double dependency_degree(int numrows, HeapTuple *rows,
+ ExprInfo *exprs, int k,
+ AttrNumber *dependency, VacAttrStats **stats,
+ Bitmapset *attrs);
static bool dependency_is_fully_matched(MVDependency *dependency,
Bitmapset *attnums);
static bool dependency_is_compatible_clause(Node *clause, Index relid,
AttrNumber *attnum);
+static bool dependency_is_compatible_expression(Node *clause, Index relid,
+ List *statlist, Node **expr);
static MVDependency *find_strongest_dependency(MVDependencies **dependencies,
- int ndependencies,
- Bitmapset *attnums);
+ int ndependencies, Bitmapset *attnums);
static Selectivity clauselist_apply_dependencies(PlannerInfo *root, List *clauses,
int varRelid, JoinType jointype,
SpecialJoinInfo *sjinfo,
@@ -219,8 +222,9 @@ DependencyGenerator_next(DependencyGenerator state)
* the last one.
*/
static double
-dependency_degree(int numrows, HeapTuple *rows, int k, AttrNumber *dependency,
- VacAttrStats **stats, Bitmapset *attrs)
+dependency_degree(int numrows, HeapTuple *rows, ExprInfo *exprs, int k,
+ AttrNumber *dependency, VacAttrStats **stats,
+ Bitmapset *attrs)
{
int i,
nitems;
@@ -289,8 +293,8 @@ dependency_degree(int numrows, HeapTuple *rows, int k, AttrNumber *dependency,
* descriptor. For now that assumption holds, but it might change in the
* future for example if we support statistics on multiple tables.
*/
- items = build_sorted_items(numrows, &nitems, rows, stats[0]->tupDesc,
- mss, k, attnums_dep);
+ items = build_sorted_items(numrows, &nitems, rows, exprs,
+ stats[0]->tupDesc, mss, k, attnums_dep);
/*
* Walk through the sorted array, split it into rows according to the
@@ -360,7 +364,8 @@ dependency_degree(int numrows, HeapTuple *rows, int k, AttrNumber *dependency,
* (c) -> b
*/
MVDependencies *
-statext_dependencies_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
+statext_dependencies_build(int numrows, HeapTuple *rows,
+ ExprInfo *exprs, Bitmapset *attrs,
VacAttrStats **stats)
{
int i,
@@ -371,6 +376,9 @@ statext_dependencies_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
/* result */
MVDependencies *dependencies = NULL;
+ /* treat expressions as special attributes with high attnums */
+ attrs = add_expressions_to_attributes(attrs, exprs->nexprs);
+
/*
* Transform the bms into an array, to make accessing i-th member easier.
*/
@@ -398,7 +406,8 @@ statext_dependencies_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
MVDependency *d;
/* compute how valid the dependency seems */
- degree = dependency_degree(numrows, rows, k, dependency, stats, attrs);
+ degree = dependency_degree(numrows, rows, exprs, k, dependency,
+ stats, attrs);
/*
* if the dependency seems entirely invalid, don't store it
@@ -441,6 +450,8 @@ statext_dependencies_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
DependencyGenerator_free(DependencyGenerator);
}
+ pfree(attrs);
+
return dependencies;
}
@@ -639,7 +650,7 @@ statext_dependencies_load(Oid mvoid)
Anum_pg_statistic_ext_data_stxddependencies, &isnull);
if (isnull)
elog(ERROR,
- "requested statistic kind \"%c\" is not yet built for statistics object %u",
+ "requested statistics kind \"%c\" is not yet built for statistics object %u",
STATS_EXT_DEPENDENCIES, mvoid);
result = statext_dependencies_deserialize(DatumGetByteaPP(deps));
@@ -1157,6 +1168,134 @@ clauselist_apply_dependencies(PlannerInfo *root, List *clauses,
return s1;
}
+/*
+ * dependency_is_compatible_expression
+ * Determines if the expression is compatible with functional dependencies
+ *
+ * Similar to dependency_is_compatible_clause, but doesn't enforce that the
+ * expression is a simple Var. OTOH we check that there's at least one
+ * statistics object matching the expression.
+ */
+static bool
+dependency_is_compatible_expression(Node *clause, Index relid, List *statlist, Node **expr)
+{
+ List *vars;
+ ListCell *lc, *lc2;
+
+ RestrictInfo *rinfo = (RestrictInfo *) clause;
+ Node *clause_expr;
+
+ if (!IsA(rinfo, RestrictInfo))
+ return false;
+
+ /* Pseudoconstants are not interesting (they couldn't contain a Var) */
+ if (rinfo->pseudoconstant)
+ return false;
+
+ /* Clauses referencing multiple, or no, varnos are incompatible */
+ if (bms_membership(rinfo->clause_relids) != BMS_SINGLETON)
+ return false;
+
+ if (is_opclause(rinfo->clause))
+ {
+ /* If it's an opclause, check for Var = Const or Const = Var. */
+ OpExpr *expr = (OpExpr *) rinfo->clause;
+
+ /* Only expressions with two arguments are candidates. */
+ if (list_length(expr->args) != 2)
+ return false;
+
+ /* Make sure non-selected argument is a pseudoconstant. */
+ if (is_pseudo_constant_clause(lsecond(expr->args)))
+ clause_expr = linitial(expr->args);
+ else if (is_pseudo_constant_clause(linitial(expr->args)))
+ clause_expr = lsecond(expr->args);
+ else
+ return false;
+
+ /*
+ * If it's not an "=" operator, just ignore the clause, as it's not
+ * compatible with functional dependencies.
+ *
+ * This uses the function for estimating selectivity, not the operator
+ * directly (a bit awkward, but well ...).
+ *
+ * XXX this is pretty dubious; probably it'd be better to check btree
+ * or hash opclass membership, so as not to be fooled by custom
+ * selectivity functions, and to be more consistent with decisions
+ * elsewhere in the planner.
+ */
+ if (get_oprrest(expr->opno) != F_EQSEL)
+ return false;
+
+ /* OK to proceed with checking "var" */
+ }
+ else if (is_notclause(rinfo->clause))
+ {
+ /*
+ * "NOT x" can be interpreted as "x = false", so get the argument and
+ * proceed with seeing if it's a suitable Var.
+ */
+ clause_expr = (Node *) get_notclausearg(rinfo->clause);
+ }
+ else
+ {
+ /*
+ * A boolean expression "x" can be interpreted as "x = true", so
+ * proceed with seeing if it's a suitable Var.
+ */
+ clause_expr = (Node *) rinfo->clause;
+ }
+
+ /*
+ * We may ignore any RelabelType node above the operand. (There won't be
+ * more than one, since eval_const_expressions has been applied already.)
+ */
+ if (IsA(clause_expr, RelabelType))
+ clause_expr = (Node *) ((RelabelType *) clause_expr)->arg;
+
+ vars = pull_var_clause(clause_expr, 0);
+
+ foreach (lc, vars)
+ {
+ Var *var = (Var *) lfirst(lc);
+
+ /* Ensure Var is from the correct relation */
+ if (var->varno != relid)
+ return false;
+
+ /* We also better ensure the Var is from the current level */
+ if (var->varlevelsup != 0)
+ return false;
+
+ /* Also ignore system attributes (we don't allow stats on those) */
+ if (!AttrNumberIsForUserDefinedAttr(var->varattno))
+ return false;
+ }
+
+ foreach (lc, statlist)
+ {
+ StatisticExtInfo *info = (StatisticExtInfo *) lfirst(lc);
+
+ /* ignore stats without dependencies */
+ if (info->kind != STATS_EXT_DEPENDENCIES)
+ continue;
+
+ foreach (lc2, info->exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc2);
+
+ if (equal(clause_expr, stat_expr))
+ {
+ *expr = stat_expr;
+ return true;
+ }
+ }
+ }
+
+ return false;
+}
+
/*
* dependencies_clauselist_selectivity
* Return the estimated selectivity of (a subset of) the given clauses
@@ -1205,6 +1344,10 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
int ndependencies;
int i;
+ /* unique expressions */
+ Node **unique_exprs;
+ int unique_exprs_cnt;
+
/* check if there's any stats that might be useful for us. */
if (!has_stats_of_kind(rel->statlist, STATS_EXT_DEPENDENCIES))
return 1.0;
@@ -1212,6 +1355,14 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
list_attnums = (AttrNumber *) palloc(sizeof(AttrNumber) *
list_length(clauses));
+ /*
+ * We allocate space as if every clause was a unique expression, although
+ * that's probably overkill. Some will be simple column references that
+ * we'll translate to attnums, and there might be duplicates.
+ */
+ unique_exprs = (Node **) palloc(sizeof(Node *) * list_length(clauses));
+ unique_exprs_cnt = 0;
+
/*
* Pre-process the clauses list to extract the attnums seen in each item.
* We need to determine if there's any clauses which will be useful for
@@ -1222,29 +1373,76 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
*
* We also skip clauses that we already estimated using different types of
* statistics (we treat them as incompatible).
+ *
+ * For expressions, we generate attnums higher than MaxHeapAttributeNumber
+ * so that we can work with attnums only.
*/
listidx = 0;
foreach(l, clauses)
{
Node *clause = (Node *) lfirst(l);
AttrNumber attnum;
+ Node *expr = NULL;
- if (!bms_is_member(listidx, *estimatedclauses) &&
- dependency_is_compatible_clause(clause, rel->relid, &attnum))
+ /* ignore clause by default */
+ list_attnums[listidx] = InvalidAttrNumber;
+
+ if (!bms_is_member(listidx, *estimatedclauses))
{
- list_attnums[listidx] = attnum;
- clauses_attnums = bms_add_member(clauses_attnums, attnum);
+ /*
+ * If it's a simple column refrence, just extract the attnum. If
+ * it's an expression, make sure it's not a duplicate and assign
+ * a special attnum to it (higher than any regular value).
+ */
+ if (dependency_is_compatible_clause(clause, rel->relid, &attnum))
+ {
+ list_attnums[listidx] = attnum;
+ clauses_attnums = bms_add_member(clauses_attnums, attnum);
+ }
+ else if (dependency_is_compatible_expression(clause, rel->relid,
+ rel->statlist,
+ &expr))
+ {
+ /* special attnum assigned to this expression */
+ attnum = InvalidAttrNumber;
+
+ Assert(expr != NULL);
+
+ /* If the expression is duplicate, use the same attnum. */
+ for (i = 0; i < unique_exprs_cnt; i++)
+ {
+ if (equal(unique_exprs[i], expr))
+ {
+ attnum = EXPRESSION_ATTNUM(i);
+ break;
+ }
+ }
+
+ /* not found in the list, so add it */
+ if (attnum == InvalidAttrNumber)
+ {
+ attnum = EXPRESSION_ATTNUM(unique_exprs_cnt);
+ unique_exprs[unique_exprs_cnt++] = expr;
+
+ /* shouldn't have seen this attnum yet */
+ Assert(!bms_is_member(attnum, clauses_attnums));
+
+ /* we may add the attnum repeatedly to clauses_attnums */
+ clauses_attnums = bms_add_member(clauses_attnums, attnum);
+ }
+
+ /* remember which attnum was assigned to this clause */
+ list_attnums[listidx] = attnum;
+ }
}
- else
- list_attnums[listidx] = InvalidAttrNumber;
listidx++;
}
/*
- * If there's not at least two distinct attnums then reject the whole list
- * of clauses. We must return 1.0 so the calling function's selectivity is
- * unaffected.
+ * If there's not at least two distinct attnums and expressions, then
+ * reject the whole list of clauses. We must return 1.0 so the calling
+ * function's selectivity is unaffected.
*/
if (bms_membership(clauses_attnums) != BMS_MULTIPLE)
{
@@ -1273,25 +1471,138 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
{
StatisticExtInfo *stat = (StatisticExtInfo *) lfirst(l);
Bitmapset *matched;
- BMS_Membership membership;
+ int nmatched;
+ int nexprs;
+ MVDependencies *deps;
/* skip statistics that are not of the correct type */
if (stat->kind != STATS_EXT_DEPENDENCIES)
continue;
+ /* count matching simple clauses */
matched = bms_intersect(clauses_attnums, stat->keys);
- membership = bms_membership(matched);
+ nmatched = bms_num_members(matched);
bms_free(matched);
- /* skip objects matching fewer than two attributes from clauses */
- if (membership != BMS_MULTIPLE)
+ /* count matching expressions */
+ nexprs = 0;
+ for (i = 0; i < unique_exprs_cnt; i++)
+ {
+ ListCell *lc;
+
+ foreach (lc, stat->exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc);
+
+ /* try to match it */
+ if (equal(stat_expr, unique_exprs[i]))
+ nexprs++;
+ }
+ }
+
+ /*
+ * Skip objects matching fewer than two attributes/expressions
+ * from clauses.
+ */
+ if (nmatched + nexprs < 2)
continue;
- func_dependencies[nfunc_dependencies]
- = statext_dependencies_load(stat->statOid);
+ deps = statext_dependencies_load(stat->statOid);
- total_ndeps += func_dependencies[nfunc_dependencies]->ndeps;
- nfunc_dependencies++;
+ /*
+ * The expressions may be represented by different attnums in the
+ * stats, we need to remap them to be consistent with the clauses.
+ * That will make the later steps (e.g. picking the strongest item
+ * and so on) much simpler.
+ *
+ * When we're at it, we can also remove dependencies referencing
+ * missing clauses (i.e. expressions that are not in the clauses).
+ *
+ * XXX We might also skip clauses referencing missing attnums, not
+ * just expressions.
+ */
+ if (stat->exprs)
+ {
+ int ndeps = 0;
+
+ for (i = 0; i < deps->ndeps; i++)
+ {
+ bool skip = false;
+ MVDependency *dep = deps->deps[i];
+ int j;
+
+ for (j = 0; j < dep->nattributes; j++)
+ {
+ int idx;
+ Node *expr;
+ int k;
+ AttrNumber unique_attnum = InvalidAttrNumber;
+
+ /* regular attribute, no need to remap */
+ if (dep->attributes[j] <= MaxHeapAttributeNumber)
+ continue;
+
+ /* index of the expression */
+ idx = EXPRESSION_INDEX(dep->attributes[j]);
+
+ /* make sure the expression index is valid */
+ Assert((idx >= 0) && (idx < list_length(stat->exprs)));
+
+ expr = (Node *) list_nth(stat->exprs, idx);
+
+ /* try to find the expression in the unique list */
+ for (k = 0; k < unique_exprs_cnt; k++)
+ {
+ /*
+ * found a matching unique expression, use the attnum
+ * (derived from index of the unique expression)
+ */
+ if (equal(unique_exprs[k], expr))
+ {
+ unique_attnum = EXPRESSION_ATTNUM(k);
+ break;
+ }
+ }
+
+ /*
+ * Not found a matching expression, so we can simply
+ * skip this dependency, because there's no chance it
+ * will be fully covered.
+ */
+ if (unique_attnum == InvalidAttrNumber)
+ {
+ skip = true;
+ break;
+ }
+
+ /* otherwise remap it to the new attnum */
+ dep->attributes[j] = unique_attnum;
+ }
+
+ /* if found a matching, */
+ if (!skip)
+ {
+ /* maybe we've skipped something earlier, so move it */
+ if (ndeps != i)
+ deps->deps[ndeps] = deps->deps[i];
+
+ ndeps++;
+ }
+ }
+
+ deps->ndeps = ndeps;
+ }
+
+ /*
+ * It's possible we've removed all dependencies, in which case we
+ * don't bother adding it to the list.
+ */
+ if (deps->ndeps > 0)
+ {
+ func_dependencies[nfunc_dependencies] = deps;
+ total_ndeps += deps->ndeps;
+ nfunc_dependencies++;
+ }
}
/* if no matching stats could be found then we've nothing to do */
@@ -1300,6 +1611,7 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
pfree(func_dependencies);
bms_free(clauses_attnums);
pfree(list_attnums);
+ pfree(unique_exprs);
return 1.0;
}
@@ -1347,6 +1659,7 @@ dependencies_clauselist_selectivity(PlannerInfo *root,
pfree(func_dependencies);
bms_free(clauses_attnums);
pfree(list_attnums);
+ pfree(unique_exprs);
return s1;
}
diff --git a/src/backend/statistics/extended_stats.c b/src/backend/statistics/extended_stats.c
index a030ea3653..fd6e160ff4 100644
--- a/src/backend/statistics/extended_stats.c
+++ b/src/backend/statistics/extended_stats.c
@@ -24,6 +24,7 @@
#include "catalog/pg_collation.h"
#include "catalog/pg_statistic_ext.h"
#include "catalog/pg_statistic_ext_data.h"
+#include "executor/executor.h"
#include "commands/progress.h"
#include "miscadmin.h"
#include "nodes/nodeFuncs.h"
@@ -35,13 +36,16 @@
#include "statistics/statistics.h"
#include "utils/acl.h"
#include "utils/array.h"
+#include "utils/attoptcache.h"
#include "utils/builtins.h"
+#include "utils/datum.h"
#include "utils/fmgroids.h"
#include "utils/lsyscache.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#include "utils/selfuncs.h"
#include "utils/syscache.h"
+#include "utils/typcache.h"
/*
* To avoid consuming too much memory during analysis and/or too much space
@@ -64,20 +68,37 @@ typedef struct StatExtEntry
char *schema; /* statistics object's schema */
char *name; /* statistics object's name */
Bitmapset *columns; /* attribute numbers covered by the object */
- List *types; /* 'char' list of enabled statistic kinds */
+ List *types; /* 'char' list of enabled statistics kinds */
int stattarget; /* statistics target (-1 for default) */
+ List *exprs; /* expressions */
} StatExtEntry;
static List *fetch_statentries_for_relation(Relation pg_statext, Oid relid);
-static VacAttrStats **lookup_var_attr_stats(Relation rel, Bitmapset *attrs,
+static VacAttrStats **lookup_var_attr_stats(Relation rel, Bitmapset *attrs, List *exprs,
int nvacatts, VacAttrStats **vacatts);
-static void statext_store(Oid relid,
+static void statext_store(Oid statOid,
MVNDistinct *ndistinct, MVDependencies *dependencies,
- MCVList *mcv, VacAttrStats **stats);
+ MCVList *mcv, Datum exprs, VacAttrStats **stats);
static int statext_compute_stattarget(int stattarget,
int natts, VacAttrStats **stats);
+typedef struct AnlExprData
+{
+ Node *expr; /* expression to analyze */
+ VacAttrStats *vacattrstat; /* index attrs to analyze */
+} AnlExprData;
+
+static void compute_expr_stats(Relation onerel, double totalrows,
+ AnlExprData *exprdata, int nexprs,
+ HeapTuple *rows, int numrows);
+static Datum serialize_expr_stats(AnlExprData *exprdata, int nexprs);
+static Datum expr_fetch_func(VacAttrStatsP stats, int rownum, bool *isNull);
+static AnlExprData *build_expr_data(List *exprs);
+static VacAttrStats *examine_expression(Node *expr);
+static ExprInfo *evaluate_expressions(Relation rel, List *exprs,
+ int numrows, HeapTuple *rows);
+
/*
* Compute requested extended stats, using the rows sampled for the plain
* (single-column) stats.
@@ -92,7 +113,7 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
{
Relation pg_stext;
ListCell *lc;
- List *stats;
+ List *statslist;
MemoryContext cxt;
MemoryContext oldcxt;
int64 ext_cnt;
@@ -103,10 +124,10 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
oldcxt = MemoryContextSwitchTo(cxt);
pg_stext = table_open(StatisticExtRelationId, RowExclusiveLock);
- stats = fetch_statentries_for_relation(pg_stext, RelationGetRelid(onerel));
+ statslist = fetch_statentries_for_relation(pg_stext, RelationGetRelid(onerel));
/* report this phase */
- if (stats != NIL)
+ if (statslist != NIL)
{
const int index[] = {
PROGRESS_ANALYZE_PHASE,
@@ -114,28 +135,30 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
};
const int64 val[] = {
PROGRESS_ANALYZE_PHASE_COMPUTE_EXT_STATS,
- list_length(stats)
+ list_length(statslist)
};
pgstat_progress_update_multi_param(2, index, val);
}
ext_cnt = 0;
- foreach(lc, stats)
+ foreach(lc, statslist)
{
StatExtEntry *stat = (StatExtEntry *) lfirst(lc);
MVNDistinct *ndistinct = NULL;
MVDependencies *dependencies = NULL;
MCVList *mcv = NULL;
+ Datum exprstats = (Datum) 0;
VacAttrStats **stats;
ListCell *lc2;
int stattarget;
+ ExprInfo *exprs;
/*
* Check if we can build these stats based on the column analyzed. If
* not, report this fact (except in autovacuum) and move on.
*/
- stats = lookup_var_attr_stats(onerel, stat->columns,
+ stats = lookup_var_attr_stats(onerel, stat->columns, stat->exprs,
natts, vacattrstats);
if (!stats)
{
@@ -150,10 +173,6 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
continue;
}
- /* check allowed number of dimensions */
- Assert(bms_num_members(stat->columns) >= 2 &&
- bms_num_members(stat->columns) <= STATS_MAX_DIMENSIONS);
-
/* compute statistics target for this statistics */
stattarget = statext_compute_stattarget(stat->stattarget,
bms_num_members(stat->columns),
@@ -167,6 +186,9 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
if (stattarget == 0)
continue;
+ /* evaluate expressions (if the statistics has any) */
+ exprs = evaluate_expressions(onerel, stat->exprs, numrows, rows);
+
/* compute statistic of each requested type */
foreach(lc2, stat->types)
{
@@ -174,21 +196,43 @@ BuildRelationExtStatistics(Relation onerel, double totalrows,
if (t == STATS_EXT_NDISTINCT)
ndistinct = statext_ndistinct_build(totalrows, numrows, rows,
- stat->columns, stats);
+ exprs, stat->columns,
+ stats);
else if (t == STATS_EXT_DEPENDENCIES)
dependencies = statext_dependencies_build(numrows, rows,
- stat->columns, stats);
+ exprs, stat->columns,
+ stats);
else if (t == STATS_EXT_MCV)
- mcv = statext_mcv_build(numrows, rows, stat->columns, stats,
- totalrows, stattarget);
+ mcv = statext_mcv_build(numrows, rows, exprs, stat->columns,
+ stats, totalrows, stattarget);
+ else if (t == STATS_EXT_EXPRESSIONS)
+ {
+ AnlExprData *exprdata;
+ int nexprs;
+
+ /* should not happen, thanks to checks when defining stats */
+ if (!stat->exprs)
+ elog(ERROR, "requested expression stats, but there are no expressions");
+
+ exprdata = build_expr_data(stat->exprs);
+ nexprs = list_length(stat->exprs);
+
+ compute_expr_stats(onerel, totalrows,
+ exprdata, nexprs,
+ rows, numrows);
+
+ exprstats = serialize_expr_stats(exprdata, nexprs);
+ }
}
/* store the statistics in the catalog */
- statext_store(stat->statOid, ndistinct, dependencies, mcv, stats);
+ statext_store(stat->statOid, ndistinct, dependencies, mcv, exprstats, stats);
/* for reporting progress */
pgstat_progress_update_param(PROGRESS_ANALYZE_EXT_STATS_COMPUTED,
++ext_cnt);
+
+ pfree(exprs);
}
table_close(pg_stext, RowExclusiveLock);
@@ -221,6 +265,13 @@ ComputeExtStatisticsRows(Relation onerel,
MemoryContext oldcxt;
int result = 0;
+ /*
+ * When there are no columns to analyze, just return 0. That's enough
+ * for the callers to not build anything.
+ */
+ if (!natts)
+ return 0;
+
cxt = AllocSetContextCreate(CurrentMemoryContext,
"ComputeExtStatisticsRows",
ALLOCSET_DEFAULT_SIZES);
@@ -241,7 +292,7 @@ ComputeExtStatisticsRows(Relation onerel,
* analyzed. If not, ignore it (don't report anything, we'll do that
* during the actual build BuildRelationExtStatistics).
*/
- stats = lookup_var_attr_stats(onerel, stat->columns,
+ stats = lookup_var_attr_stats(onerel, stat->columns, stat->exprs,
natts, vacattrstats);
if (!stats)
@@ -349,6 +400,10 @@ statext_is_kind_built(HeapTuple htup, char type)
attnum = Anum_pg_statistic_ext_data_stxdmcv;
break;
+ case STATS_EXT_EXPRESSIONS:
+ attnum = Anum_pg_statistic_ext_data_stxdexpr;
+ break;
+
default:
elog(ERROR, "unexpected statistics type requested: %d", type);
}
@@ -388,6 +443,7 @@ fetch_statentries_for_relation(Relation pg_statext, Oid relid)
ArrayType *arr;
char *enabled;
Form_pg_statistic_ext staForm;
+ List *exprs = NIL;
entry = palloc0(sizeof(StatExtEntry));
staForm = (Form_pg_statistic_ext) GETSTRUCT(htup);
@@ -415,10 +471,39 @@ fetch_statentries_for_relation(Relation pg_statext, Oid relid)
{
Assert((enabled[i] == STATS_EXT_NDISTINCT) ||
(enabled[i] == STATS_EXT_DEPENDENCIES) ||
- (enabled[i] == STATS_EXT_MCV));
+ (enabled[i] == STATS_EXT_MCV) ||
+ (enabled[i] == STATS_EXT_EXPRESSIONS));
entry->types = lappend_int(entry->types, (int) enabled[i]);
}
+ /* decode expression (if any) */
+ datum = SysCacheGetAttr(STATEXTOID, htup,
+ Anum_pg_statistic_ext_stxexprs, &isnull);
+
+ if (!isnull)
+ {
+ char *exprsString;
+
+ exprsString = TextDatumGetCString(datum);
+ exprs = (List *) stringToNode(exprsString);
+
+ pfree(exprsString);
+
+ /*
+ * Run the expressions through eval_const_expressions. This is not just an
+ * optimization, but is necessary, because the planner will be comparing
+ * them to similarly-processed qual clauses, and may fail to detect valid
+ * matches without this. We must not use canonicalize_qual, however,
+ * since these aren't qual expressions.
+ */
+ exprs = (List *) eval_const_expressions(NULL, (Node *) exprs);
+
+ /* May as well fix opfuncids too */
+ fix_opfuncids((Node *) exprs);
+ }
+
+ entry->exprs = exprs;
+
result = lappend(result, entry);
}
@@ -427,6 +512,86 @@ fetch_statentries_for_relation(Relation pg_statext, Oid relid)
return result;
}
+
+/*
+ * examine_attribute -- pre-analysis of a single column
+ *
+ * Determine whether the column is analyzable; if so, create and initialize
+ * a VacAttrStats struct for it. If not, return NULL.
+ */
+static VacAttrStats *
+examine_attribute(Node *expr)
+{
+ HeapTuple typtuple;
+ VacAttrStats *stats;
+ int i;
+ bool ok;
+
+ /*
+ * Create the VacAttrStats struct. Note that we only have a copy of the
+ * fixed fields of the pg_attribute tuple.
+ */
+ stats = (VacAttrStats *) palloc0(sizeof(VacAttrStats));
+
+ /* fake the attribute */
+ stats->attr = (Form_pg_attribute) palloc0(ATTRIBUTE_FIXED_PART_SIZE);
+ stats->attr->attstattarget = -1;
+
+ /*
+ * When analyzing an expression index, believe the expression tree's type
+ * not the column datatype --- the latter might be the opckeytype storage
+ * type of the opclass, which is not interesting for our purposes. (Note:
+ * if we did anything with non-expression index columns, we'd need to
+ * figure out where to get the correct type info from, but for now that's
+ * not a problem.) It's not clear whether anyone will care about the
+ * typmod, but we store that too just in case.
+ */
+ stats->attrtypid = exprType(expr);
+ stats->attrtypmod = exprTypmod(expr);
+ stats->attrcollid = exprCollation(expr);
+
+ typtuple = SearchSysCacheCopy1(TYPEOID,
+ ObjectIdGetDatum(stats->attrtypid));
+ if (!HeapTupleIsValid(typtuple))
+ elog(ERROR, "cache lookup failed for type %u", stats->attrtypid);
+ stats->attrtype = (Form_pg_type) GETSTRUCT(typtuple);
+ // stats->anl_context = anl_context; /* FIXME? */
+ stats->tupattnum = InvalidAttrNumber;
+
+ /*
+ * The fields describing the stats->stavalues[n] element types default to
+ * the type of the data being analyzed, but the type-specific typanalyze
+ * function can change them if it wants to store something else.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ stats->statypid[i] = stats->attrtypid;
+ stats->statyplen[i] = stats->attrtype->typlen;
+ stats->statypbyval[i] = stats->attrtype->typbyval;
+ stats->statypalign[i] = stats->attrtype->typalign;
+ }
+
+ /*
+ * Call the type-specific typanalyze function. If none is specified, use
+ * std_typanalyze().
+ */
+ if (OidIsValid(stats->attrtype->typanalyze))
+ ok = DatumGetBool(OidFunctionCall1(stats->attrtype->typanalyze,
+ PointerGetDatum(stats)));
+ else
+ ok = std_typanalyze(stats);
+
+ if (!ok || stats->compute_stats == NULL || stats->minrows <= 0)
+ {
+ heap_freetuple(typtuple);
+ pfree(stats->attr);
+ pfree(stats);
+ return NULL;
+ }
+
+ return stats;
+}
+
/*
* Using 'vacatts' of size 'nvacatts' as input data, return a newly built
* VacAttrStats array which includes only the items corresponding to
@@ -435,15 +600,18 @@ fetch_statentries_for_relation(Relation pg_statext, Oid relid)
* to the caller that the stats should not be built.
*/
static VacAttrStats **
-lookup_var_attr_stats(Relation rel, Bitmapset *attrs,
+lookup_var_attr_stats(Relation rel, Bitmapset *attrs, List *exprs,
int nvacatts, VacAttrStats **vacatts)
{
int i = 0;
int x = -1;
+ int natts;
VacAttrStats **stats;
+ ListCell *lc;
- stats = (VacAttrStats **)
- palloc(bms_num_members(attrs) * sizeof(VacAttrStats *));
+ natts = bms_num_members(attrs) + list_length(exprs);
+
+ stats = (VacAttrStats **) palloc(natts * sizeof(VacAttrStats *));
/* lookup VacAttrStats info for the requested columns (same attnum) */
while ((x = bms_next_member(attrs, x)) >= 0)
@@ -480,6 +648,24 @@ lookup_var_attr_stats(Relation rel, Bitmapset *attrs,
i++;
}
+ /* also add info for expressions */
+ foreach (lc, exprs)
+ {
+ Node *expr = (Node *) lfirst(lc);
+
+ stats[i] = examine_attribute(expr);
+
+ /*
+ * FIXME We need tuple descriptor later, and we just grab it from
+ * stats[0]->tupDesc (see e.g. statext_mcv_build). But as coded
+ * examine_attribute does not set that, so just grab it from the
+ * first vacatts element.
+ */
+ stats[i]->tupDesc = vacatts[0]->tupDesc;
+
+ i++;
+ }
+
return stats;
}
@@ -491,7 +677,7 @@ lookup_var_attr_stats(Relation rel, Bitmapset *attrs,
static void
statext_store(Oid statOid,
MVNDistinct *ndistinct, MVDependencies *dependencies,
- MCVList *mcv, VacAttrStats **stats)
+ MCVList *mcv, Datum exprs, VacAttrStats **stats)
{
Relation pg_stextdata;
HeapTuple stup,
@@ -532,11 +718,17 @@ statext_store(Oid statOid,
nulls[Anum_pg_statistic_ext_data_stxdmcv - 1] = (data == NULL);
values[Anum_pg_statistic_ext_data_stxdmcv - 1] = PointerGetDatum(data);
}
+ if (exprs != (Datum) 0)
+ {
+ nulls[Anum_pg_statistic_ext_data_stxdexpr - 1] = false;
+ values[Anum_pg_statistic_ext_data_stxdexpr - 1] = exprs;
+ }
/* always replace the value (either by bytea or NULL) */
replaces[Anum_pg_statistic_ext_data_stxdndistinct - 1] = true;
replaces[Anum_pg_statistic_ext_data_stxddependencies - 1] = true;
replaces[Anum_pg_statistic_ext_data_stxdmcv - 1] = true;
+ replaces[Anum_pg_statistic_ext_data_stxdexpr - 1] = true;
/* there should already be a pg_statistic_ext_data tuple */
oldtup = SearchSysCache1(STATEXTDATASTXOID, ObjectIdGetDatum(statOid));
@@ -741,8 +933,9 @@ build_attnums_array(Bitmapset *attrs, int *numattrs)
* can simply pfree the return value to release all of it.
*/
SortItem *
-build_sorted_items(int numrows, int *nitems, HeapTuple *rows, TupleDesc tdesc,
- MultiSortSupport mss, int numattrs, AttrNumber *attnums)
+build_sorted_items(int numrows, int *nitems, HeapTuple *rows, ExprInfo *exprs,
+ TupleDesc tdesc, MultiSortSupport mss,
+ int numattrs, AttrNumber *attnums)
{
int i,
j,
@@ -789,8 +982,24 @@ build_sorted_items(int numrows, int *nitems, HeapTuple *rows, TupleDesc tdesc,
{
Datum value;
bool isnull;
+ int attlen;
- value = heap_getattr(rows[i], attnums[j], tdesc, &isnull);
+ if (attnums[j] <= MaxHeapAttributeNumber)
+ {
+ value = heap_getattr(rows[i], attnums[j], tdesc, &isnull);
+ attlen = TupleDescAttr(tdesc, attnums[j] - 1)->attlen;
+ }
+ else
+ {
+ int idx = EXPRESSION_INDEX(attnums[j]);
+
+ Assert((idx >= 0) && (idx < exprs->nexprs));
+
+ value = exprs->values[idx][i];
+ isnull = exprs->nulls[idx][i];
+
+ attlen = get_typlen(exprs->types[idx]);
+ }
/*
* If this is a varlena value, check if it's too wide and if yes
@@ -801,8 +1010,7 @@ build_sorted_items(int numrows, int *nitems, HeapTuple *rows, TupleDesc tdesc,
* on the assumption that those are small (below WIDTH_THRESHOLD)
* and will be discarded at the end of analyze.
*/
- if ((!isnull) &&
- (TupleDescAttr(tdesc, attnums[j] - 1)->attlen == -1))
+ if ((!isnull) && (attlen == -1))
{
if (toast_raw_datum_size(value) > WIDTH_THRESHOLD)
{
@@ -881,7 +1089,8 @@ has_stats_of_kind(List *stats, char requiredkind)
*/
StatisticExtInfo *
choose_best_statistics(List *stats, char requiredkind,
- Bitmapset **clause_attnums, int nclauses)
+ Bitmapset **clause_attnums, List **clause_exprs,
+ int nclauses)
{
ListCell *lc;
StatisticExtInfo *best_match = NULL;
@@ -894,6 +1103,7 @@ choose_best_statistics(List *stats, char requiredkind,
StatisticExtInfo *info = (StatisticExtInfo *) lfirst(lc);
Bitmapset *matched = NULL;
int num_matched;
+ int num_matched_exprs;
int numkeys;
/* skip statistics that are not of the correct type */
@@ -920,6 +1130,38 @@ choose_best_statistics(List *stats, char requiredkind,
num_matched = bms_num_members(matched);
bms_free(matched);
+ /*
+ * Collect expressions in remaining (unestimated) expressions, covered
+ * by an expression in this statistic object.
+ */
+ num_matched_exprs = 0;
+ for (i = 0; i < nclauses; i++)
+ {
+ ListCell *lc3;
+
+ /* ignore incompatible/estimated expressions */
+ if (!clause_exprs[i])
+ continue;
+
+ /* ignore expressions that are not covered by this object */
+ foreach (lc3, clause_exprs[i])
+ {
+ ListCell *lc2;
+ Node *expr = (Node *) lfirst(lc3);
+
+ foreach(lc2, info->exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc2);
+
+ if (equal(expr, stat_expr))
+ {
+ num_matched_exprs++;
+ break;
+ }
+ }
+ }
+ }
+
/*
* save the actual number of keys in the stats so that we can choose
* the narrowest stats with the most matching keys.
@@ -931,11 +1173,12 @@ choose_best_statistics(List *stats, char requiredkind,
* when it matches the same number of attributes but these stats have
* fewer keys than any previous match.
*/
- if (num_matched > best_num_matched ||
- (num_matched == best_num_matched && numkeys < best_match_keys))
+ if (num_matched + num_matched_exprs > best_num_matched ||
+ ((num_matched + num_matched_exprs) == best_num_matched &&
+ numkeys < best_match_keys))
{
best_match = info;
- best_num_matched = num_matched;
+ best_num_matched = num_matched + num_matched_exprs;
best_match_keys = numkeys;
}
}
@@ -994,7 +1237,7 @@ statext_is_compatible_clause_internal(PlannerInfo *root, Node *clause,
return false;
/* Check if the expression has the right shape (one Var, one Const) */
- if (!examine_clause_args(expr->args, &var, NULL, NULL))
+ if (!examine_opclause_expression(expr, &var, NULL, NULL))
return false;
/*
@@ -1150,6 +1393,214 @@ statext_is_compatible_clause_internal(PlannerInfo *root, Node *clause,
return false;
}
+/*
+ * statext_extract_expression_internal
+ * Extract parts of an expressions to match against extended stats.
+ *
+ * Given an expression, decompose it into "parts" that will be analyzed and
+ * matched against extended statistics. If the expression is not considered
+ * compatible (supported by extended statistics), this returns NIL.
+ *
+ * There's a certain amount of ambiguity, because some expressions may be
+ * split into parts in multiple ways. For example, consider expression
+ *
+ * (a + b) = 1
+ *
+ * which may be either considered as a single boolean expression, or it may
+ * be split into expression (a + b) and a constant. So this might return
+ * either ((a+b)=1) or (a+b) as valid expressions, but this does affect
+ * matching to extended statistics, because the expressions have to match
+ * the definition exactly. So ((a+b)=1) would match statistics defined as
+ *
+ * CREATE STATISTICS s ON ((a+b) = 1) FROM t;
+ *
+ * but not
+ *
+ * CREATE STATISTICS s ON (a+b) FROM t;
+ *
+ * which might be a bit confusing. We might enhance this to track those
+ * alternative decompositions somehow, and then modify the matching to
+ * extended statistics. But it seems non-trivial, because the AND/OR
+ * clauses make it "recursive".
+ *
+ * in which expressions might be extracted.
+ */
+static List *
+statext_extract_expression_internal(PlannerInfo *root, Node *clause, Index relid)
+{
+ /* Look inside any binary-compatible relabeling (as in examine_variable) */
+ if (IsA(clause, RelabelType))
+ clause = (Node *) ((RelabelType *) clause)->arg;
+
+ /* plain Var references (boolean Vars or recursive checks) */
+ if (IsA(clause, Var))
+ {
+ Var *var = (Var *) clause;
+
+ /* Ensure var is from the correct relation */
+ if (var->varno != relid)
+ return NIL;
+
+ /* we also better ensure the Var is from the current level */
+ if (var->varlevelsup > 0)
+ return NIL;
+
+ /* Also skip system attributes (we don't allow stats on those). */
+ if (!AttrNumberIsForUserDefinedAttr(var->varattno))
+ return NIL;
+
+ return list_make1(clause);
+ }
+
+ /* (Var op Const) or (Const op Var) */
+ if (is_opclause(clause))
+ {
+ RangeTblEntry *rte = root->simple_rte_array[relid];
+ OpExpr *expr = (OpExpr *) clause;
+ Node *expr2 = NULL;
+
+ /* Only expressions with two arguments are considered compatible. */
+ if (list_length(expr->args) != 2)
+ return NIL;
+
+ /* Check if the expression has the right shape (one Expr, one Const) */
+ if (!examine_opclause_expression2(expr, &expr2, NULL, NULL))
+ return NIL;
+
+ /*
+ * If it's not one of the supported operators ("=", "<", ">", etc.),
+ * just ignore the clause, as it's not compatible with MCV lists.
+ *
+ * This uses the function for estimating selectivity, not the operator
+ * directly (a bit awkward, but well ...).
+ */
+ switch (get_oprrest(expr->opno))
+ {
+ case F_EQSEL:
+ case F_NEQSEL:
+ case F_SCALARLTSEL:
+ case F_SCALARLESEL:
+ case F_SCALARGTSEL:
+ case F_SCALARGESEL:
+ /* supported, will continue with inspection of the Var */
+ break;
+
+ default:
+ /* other estimators are considered unknown/unsupported */
+ return NIL;
+ }
+
+ /*
+ * If there are any securityQuals on the RTE from security barrier
+ * views or RLS policies, then the user may not have access to all the
+ * table's data, and we must check that the operator is leak-proof.
+ *
+ * If the operator is leaky, then we must ignore this clause for the
+ * purposes of estimating with MCV lists, otherwise the operator might
+ * reveal values from the MCV list that the user doesn't have
+ * permission to see.
+ */
+ if (rte->securityQuals != NIL &&
+ !get_func_leakproof(get_opcode(expr->opno)))
+ return NIL;
+
+ return list_make1(expr2);
+ }
+
+ if (IsA(clause, ScalarArrayOpExpr))
+ {
+ RangeTblEntry *rte = root->simple_rte_array[relid];
+ ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) clause;
+ Node *expr2 = NULL;
+
+ /* Only expressions with two arguments are considered compatible. */
+ if (list_length(expr->args) != 2)
+ return NIL;
+
+ /* Check if the expression has the right shape (one Expr, one Const) */
+ if (!examine_clause_args2(expr->args, &expr2, NULL, NULL))
+ return NIL;
+
+ /*
+ * If there are any securityQuals on the RTE from security barrier
+ * views or RLS policies, then the user may not have access to all the
+ * table's data, and we must check that the operator is leak-proof.
+ *
+ * If the operator is leaky, then we must ignore this clause for the
+ * purposes of estimating with MCV lists, otherwise the operator might
+ * reveal values from the MCV list that the user doesn't have
+ * permission to see.
+ */
+ if (rte->securityQuals != NIL &&
+ !get_func_leakproof(get_opcode(expr->opno)))
+ return NIL;
+
+ return list_make1(expr2);
+ }
+
+ /* AND/OR/NOT clause */
+ if (is_andclause(clause) ||
+ is_orclause(clause) ||
+ is_notclause(clause))
+ {
+ /*
+ * AND/OR/NOT-clauses are supported if all sub-clauses are supported
+ *
+ * Perhaps we could improve this by handling mixed cases, when some of
+ * the clauses are supported and some are not. Selectivity for the
+ * supported subclauses would be computed using extended statistics,
+ * and the remaining clauses would be estimated using the traditional
+ * algorithm (product of selectivities).
+ *
+ * It however seems overly complex, and in a way we already do that
+ * because if we reject the whole clause as unsupported here, it will
+ * be eventually passed to clauselist_selectivity() which does exactly
+ * this (split into supported/unsupported clauses etc).
+ */
+ BoolExpr *expr = (BoolExpr *) clause;
+ ListCell *lc;
+ List *exprs = NIL;
+
+ foreach(lc, expr->args)
+ {
+ List *tmp;
+
+ /*
+ * Had we found incompatible clause in the arguments, treat the
+ * whole clause as incompatible.
+ */
+ tmp = statext_extract_expression_internal(root,
+ (Node *) lfirst(lc),
+ relid);
+
+ if (!tmp)
+ return NIL;
+
+ exprs = list_concat(exprs, tmp);
+ }
+
+ return exprs;
+ }
+
+ /* Var IS NULL */
+ if (IsA(clause, NullTest))
+ {
+ NullTest *nt = (NullTest *) clause;
+
+ /*
+ * Only simple (Var IS NULL) expressions supported for now. Maybe we
+ * could use examine_variable to fix this?
+ */
+ if (!IsA(nt->arg, Var))
+ return NIL;
+
+ return statext_extract_expression_internal(root, (Node *) (nt->arg),
+ relid);
+ }
+
+ return NIL;
+}
+
/*
* statext_is_compatible_clause
* Determines if the clause is compatible with MCV lists.
@@ -1163,6 +1614,8 @@ statext_is_compatible_clause_internal(PlannerInfo *root, Node *clause,
*
* (c) combinations using AND/OR/NOT
*
+ * (d) ScalarArrayOpExprs of the form (Var op ANY (array)) or (Var op ALL (array))
+ *
* In the future, the range of supported clauses may be expanded to more
* complex cases, for example (Var op Var).
*/
@@ -1250,13 +1703,108 @@ statext_is_compatible_clause(PlannerInfo *root, Node *clause, Index relid,
}
/*
- * statext_mcv_clauselist_selectivity
- * Estimate clauses using the best multi-column statistics.
+ * statext_extract_expression
+ * Determines if the clause is compatible with extended statistics.
*
- * Applies available extended (multi-column) statistics on a table. There may
- * be multiple applicable statistics (with respect to the clauses), in which
- * case we use greedy approach. In each round we select the best statistic on
- * a table (measured by the number of attributes extracted from the clauses
+ * Currently, we only support three types of clauses:
+ *
+ * (a) OpExprs of the form (Var op Const), or (Const op Var), where the op
+ * is one of ("=", "<", ">", ">=", "<=")
+ *
+ * (b) (Var IS [NOT] NULL)
+ *
+ * (c) combinations using AND/OR/NOT
+ *
+ * (d) ScalarArrayOpExprs of the form (Var op ANY (array)) or (Var op ALL (array))
+ *
+ * In the future, the range of supported clauses may be expanded to more
+ * complex cases, for example (Var op Var).
+ */
+static List *
+statext_extract_expression(PlannerInfo *root, Node *clause, Index relid)
+{
+ RestrictInfo *rinfo = (RestrictInfo *) clause;
+ RangeTblEntry *rte = root->simple_rte_array[relid];
+ List *exprs;
+ Oid userid;
+
+ if (!IsA(rinfo, RestrictInfo))
+ return NIL;
+
+ /* Pseudoconstants are not really interesting here. */
+ if (rinfo->pseudoconstant)
+ return NIL;
+
+ /* clauses referencing multiple varnos are incompatible */
+ if (bms_membership(rinfo->clause_relids) != BMS_SINGLETON)
+ return NIL;
+
+ /* Check the clause and extract expressions it's composed of. */
+ exprs = statext_extract_expression_internal(root, (Node *) rinfo->clause, relid);
+
+ /*
+ * If there are no potentially interesting expressions (supported by
+ * extended statistics), we're done;
+ */
+ if (!exprs)
+ return NIL;
+
+ /*
+ * Check that the user has permission to read all these attributes. Use
+ * checkAsUser if it's set, in case we're accessing the table via a view.
+ */
+ userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
+
+ if (pg_class_aclcheck(rte->relid, userid, ACL_SELECT) != ACLCHECK_OK)
+ {
+ Bitmapset *attnums = NULL;
+
+ /* Extract all attribute numbers from the expressions. */
+ pull_varattnos((Node *) exprs, relid, &attnums);
+
+ /* Don't have table privilege, must check individual columns */
+ if (bms_is_member(InvalidAttrNumber, attnums))
+ {
+ /* Have a whole-row reference, must have access to all columns */
+ if (pg_attribute_aclcheck_all(rte->relid, userid, ACL_SELECT,
+ ACLMASK_ALL) != ACLCHECK_OK)
+ return NIL;
+ }
+ else
+ {
+ /* Check the columns referenced by the clause */
+ int attnum = -1;
+
+ while ((attnum = bms_next_member(attnums, attnum)) >= 0)
+ {
+ AttrNumber tmp;
+
+ /* Adjust for system attributes (offset for bitmap). */
+ tmp = attnum + FirstLowInvalidHeapAttributeNumber;
+
+ /* Ignore system attributes, those can't have statistics. */
+ if (!AttrNumberIsForUserDefinedAttr(tmp))
+ return NIL;
+
+ if (pg_attribute_aclcheck(rte->relid, tmp, userid,
+ ACL_SELECT) != ACLCHECK_OK)
+ return NIL;
+ }
+ }
+ }
+
+ /* If we reach here, the clause is OK */
+ return exprs;
+}
+
+/*
+ * statext_mcv_clauselist_selectivity
+ * Estimate clauses using the best multi-column statistics.
+ *
+ * Applies available extended (multi-column) statistics on a table. There may
+ * be multiple applicable statistics (with respect to the clauses), in which
+ * case we use greedy approach. In each round we select the best statistic on
+ * a table (measured by the number of attributes extracted from the clauses
* and covered by it), and compute the selectivity for the supplied clauses.
* We repeat this process with the remaining clauses (if any), until none of
* the available statistics can be used.
@@ -1290,7 +1838,8 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
bool is_or)
{
ListCell *l;
- Bitmapset **list_attnums;
+ Bitmapset **list_attnums; /* attnums extracted from the clause */
+ List **list_exprs; /* expressions matched to any statistic */
int listidx;
Selectivity sel = (is_or) ? 0.0 : 1.0;
@@ -1301,6 +1850,9 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
list_attnums = (Bitmapset **) palloc(sizeof(Bitmapset *) *
list_length(clauses));
+ /* expressions extracted from complex expressions */
+ list_exprs = (List **) palloc(sizeof(Node *) * list_length(clauses));
+
/*
* Pre-process the clauses list to extract the attnums seen in each item.
* We need to determine if there's any clauses which will be useful for
@@ -1318,11 +1870,100 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
Node *clause = (Node *) lfirst(l);
Bitmapset *attnums = NULL;
+ /* the clause is considered incompatible by default */
+ list_attnums[listidx] = NULL;
+
+ /* and it's also not covered exactly by the statistic */
+ list_exprs[listidx] = NULL;
+
+ /*
+ * First see if the clause is simple enough to be covered directly
+ * by the attributes. If not, see if there's at least one statistic
+ * object using the expression as-is.
+ */
if (!bms_is_member(listidx, *estimatedclauses) &&
statext_is_compatible_clause(root, clause, rel->relid, &attnums))
+ {
+ /* simple expression, covered through attnum(s) */
list_attnums[listidx] = attnums;
+ }
else
- list_attnums[listidx] = NULL;
+ {
+ ListCell *lc;
+ List *exprs;
+
+ /*
+ * XXX This is kinda dubious, because we extract the smallest
+ * clauses - e.g. from (Var op Const) we extract Var. But maybe
+ * the statistics covers larger expressions, so maybe this will
+ * skip that. For example give ((a+b) + (c+d)) it's not clear
+ * if we should extract the whole clause or some smaller parts.
+ * OTOH we need (Expr op Const) so maybe we only care about the
+ * clause as a whole?
+ */
+ exprs = statext_extract_expression(root, clause, rel->relid);
+
+ /* complex expression, search for statistic covering all parts */
+ foreach(lc, rel->statlist)
+ {
+ ListCell *le;
+ StatisticExtInfo *info = (StatisticExtInfo *) lfirst(lc);
+
+ /*
+ * Assume all parts are covered by this statistics, we'll
+ * stop if we found part that is not covered.
+ */
+ bool covered = true;
+
+ /* have we already matched the expression to a statistic? */
+ Assert(!list_exprs[listidx]);
+
+ /* no expressions in the statistic */
+ if (!info->exprs)
+ continue;
+
+ foreach(le, exprs)
+ {
+ ListCell *lc2;
+ Node *expr = (Node *) lfirst(le);
+ bool found = false;
+
+ /*
+ * Walk the expressions, see if all expressions extracted from
+ * the clause are covered by the extended statistic object.
+ */
+ foreach (lc2, info->exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc2);
+
+ if (equal(expr, stat_expr))
+ {
+ found = true;
+ break;
+ }
+ }
+
+ /* found expression not covered by the statistics, stop */
+ if (!found)
+ {
+ covered = false;
+ break;
+ }
+ }
+
+ /*
+ * OK, we found a statistics covering this clause, stop looking
+ * for another one
+ */
+ if (covered)
+ {
+ /* XXX should this add the original expression instead? */
+ list_exprs[listidx] = exprs;
+ break;
+ }
+
+ }
+ }
listidx++;
}
@@ -1336,7 +1977,8 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
/* find the best suited statistics object for these attnums */
stat = choose_best_statistics(rel->statlist, STATS_EXT_MCV,
- list_attnums, list_length(clauses));
+ list_attnums, list_exprs,
+ list_length(clauses));
/*
* if no (additional) matching stats could be found then we've nothing
@@ -1359,11 +2001,13 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
{
/*
* If the clause is compatible with the selected statistics, mark
- * it as estimated and add it to the list to estimate.
+ * it as estimated and add it to the list to estimate. It may be
+ * either a simple clause, or an expression.
*/
if (list_attnums[listidx] != NULL &&
bms_is_subset(list_attnums[listidx], stat->keys))
{
+ /* simple clause (single Var) */
if (bms_membership(list_attnums[listidx]) == BMS_SINGLETON)
simple_clauses = bms_add_member(simple_clauses,
list_length(stat_clauses));
@@ -1374,6 +2018,45 @@ statext_mcv_clauselist_selectivity(PlannerInfo *root, List *clauses, int varReli
bms_free(list_attnums[listidx]);
list_attnums[listidx] = NULL;
}
+ else if (list_exprs[listidx] != NIL)
+ {
+ /* are all parts of the expression covered by the statistic? */
+ ListCell *lc;
+ int ncovered = 0;
+
+ foreach (lc, list_exprs[listidx])
+ {
+ ListCell *lc2;
+ Node *expr = (Node *) lfirst(lc);
+ bool found = false;
+
+ foreach (lc2, stat->exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc2);
+
+ if (equal(expr, stat_expr))
+ {
+ found = true;
+ break;
+ }
+ }
+
+ /* count it as covered and continue to the next expression */
+ if (found)
+ ncovered++;
+ }
+
+ /* all parts of the expression are covered by this statistics */
+ if (ncovered == list_length(list_exprs[listidx]))
+ {
+ stat_clauses = lappend(stat_clauses, (Node *) lfirst(l));
+ *estimatedclauses = bms_add_member(*estimatedclauses, listidx);
+
+ list_free(list_exprs[listidx]);
+ list_exprs[listidx] = NULL;
+ }
+
+ }
listidx++;
}
@@ -1621,3 +2304,788 @@ examine_clause_args(List *args, Var **varp, Const **cstp, bool *varonleftp)
return true;
}
+
+bool
+examine_clause_args2(List *args, Node **exprp, Const **cstp, bool *expronleftp)
+{
+ Node *expr;
+ Const *cst;
+ bool expronleft;
+ Node *leftop,
+ *rightop;
+
+ /* enforced by statext_is_compatible_clause_internal */
+ Assert(list_length(args) == 2);
+
+ leftop = linitial(args);
+ rightop = lsecond(args);
+
+ /* strip RelabelType from either side of the expression */
+ if (IsA(leftop, RelabelType))
+ leftop = (Node *) ((RelabelType *) leftop)->arg;
+
+ if (IsA(rightop, RelabelType))
+ rightop = (Node *) ((RelabelType *) rightop)->arg;
+
+ if (IsA(rightop, Const))
+ {
+ expr = (Node *) leftop;
+ cst = (Const *) rightop;
+ expronleft = true;
+ }
+ else if (IsA(leftop, Const))
+ {
+ expr = (Node *) rightop;
+ cst = (Const *) leftop;
+ expronleft = false;
+ }
+ else
+ return false;
+
+ /* return pointers to the extracted parts if requested */
+ if (exprp)
+ *exprp = expr;
+
+ if (cstp)
+ *cstp = cst;
+
+ if (expronleftp)
+ *expronleftp = expronleft;
+
+ return true;
+}
+
+bool
+examine_opclause_expression(OpExpr *expr, Var **varp, Const **cstp, bool *varonleftp)
+{
+ Var *var;
+ Const *cst;
+ bool varonleft;
+ Node *leftop,
+ *rightop;
+
+ /* enforced by statext_is_compatible_clause_internal */
+ Assert(list_length(expr->args) == 2);
+
+ leftop = linitial(expr->args);
+ rightop = lsecond(expr->args);
+
+ /* strip RelabelType from either side of the expression */
+ if (IsA(leftop, RelabelType))
+ leftop = (Node *) ((RelabelType *) leftop)->arg;
+
+ if (IsA(rightop, RelabelType))
+ rightop = (Node *) ((RelabelType *) rightop)->arg;
+
+ if (IsA(leftop, Var) && IsA(rightop, Const))
+ {
+ var = (Var *) leftop;
+ cst = (Const *) rightop;
+ varonleft = true;
+ }
+ else if (IsA(leftop, Const) && IsA(rightop, Var))
+ {
+ var = (Var *) rightop;
+ cst = (Const *) leftop;
+ varonleft = false;
+ }
+ else
+ return false;
+
+ /* return pointers to the extracted parts if requested */
+ if (varp)
+ *varp = var;
+
+ if (cstp)
+ *cstp = cst;
+
+ if (varonleftp)
+ *varonleftp = varonleft;
+
+ return true;
+}
+
+bool
+examine_opclause_expression2(OpExpr *expr, Node **exprp, Const **cstp, bool *expronleftp)
+{
+ Node *expr2;
+ Const *cst;
+ bool expronleft;
+ Node *leftop,
+ *rightop;
+
+ /* enforced by statext_is_compatible_clause_internal */
+ Assert(list_length(expr->args) == 2);
+
+ leftop = linitial(expr->args);
+ rightop = lsecond(expr->args);
+
+ /* strip RelabelType from either side of the expression */
+ if (IsA(leftop, RelabelType))
+ leftop = (Node *) ((RelabelType *) leftop)->arg;
+
+ if (IsA(rightop, RelabelType))
+ rightop = (Node *) ((RelabelType *) rightop)->arg;
+
+ if (IsA(rightop, Const))
+ {
+ expr2 = (Node *) leftop;
+ cst = (Const *) rightop;
+ expronleft = true;
+ }
+ else if (IsA(leftop, Const))
+ {
+ expr2 = (Node *) rightop;
+ cst = (Const *) leftop;
+ expronleft = false;
+ }
+ else
+ return false;
+
+ /* return pointers to the extracted parts if requested */
+ if (exprp)
+ *exprp = expr2;
+
+ if (cstp)
+ *cstp = cst;
+
+ if (expronleftp)
+ *expronleftp = expronleft;
+
+ return true;
+}
+
+
+/*
+ * Compute statistics about expressions of a relation.
+ */
+static void
+compute_expr_stats(Relation onerel, double totalrows,
+ AnlExprData *exprdata, int nexprs,
+ HeapTuple *rows, int numrows)
+{
+ MemoryContext expr_context,
+ old_context;
+ int ind,
+ i;
+
+ expr_context = AllocSetContextCreate(CurrentMemoryContext,
+ "Analyze Expression",
+ ALLOCSET_DEFAULT_SIZES);
+ old_context = MemoryContextSwitchTo(expr_context);
+
+ for (ind = 0; ind < nexprs; ind++)
+ {
+ AnlExprData *thisdata = &exprdata[ind];
+ VacAttrStats *stats = thisdata->vacattrstat;
+ Node *expr = thisdata->expr;
+ TupleTableSlot *slot;
+ EState *estate;
+ ExprContext *econtext;
+ Datum *exprvals;
+ bool *exprnulls;
+ ExprState *exprstate;
+ int tcnt;
+
+ /* Are we still in the main context? */
+ Assert(CurrentMemoryContext == expr_context);
+
+ /*
+ * Need an EState for evaluation of expressions. Create it in
+ * the per-expression context to be sure it gets cleaned up at
+ * the bottom of the loop.
+ */
+ estate = CreateExecutorState();
+ econtext = GetPerTupleExprContext(estate);
+
+ /* Set up expression evaluation state */
+ exprstate = ExecPrepareExpr((Expr *) expr, estate);
+
+ /* Need a slot to hold the current heap tuple, too */
+ slot = MakeSingleTupleTableSlot(RelationGetDescr(onerel),
+ &TTSOpsHeapTuple);
+
+ /* Arrange for econtext's scan tuple to be the tuple under test */
+ econtext->ecxt_scantuple = slot;
+
+ /* Compute and save index expression values */
+ exprvals = (Datum *) palloc(numrows * sizeof(Datum));
+ exprnulls = (bool *) palloc(numrows * sizeof(bool));
+
+ tcnt = 0;
+ for (i = 0; i < numrows; i++)
+ {
+ Datum datum;
+ bool isnull;
+
+ /*
+ * Reset the per-tuple context each time, to reclaim any cruft
+ * left behind by evaluating the predicate or index expressions.
+ */
+ ResetExprContext(econtext);
+
+ /* Set up for expression evaluation */
+ ExecStoreHeapTuple(rows[i], slot, false);
+
+ /*
+ * Evaluate the expression. We do this in the per-tuple context
+ * so as not to leak memory, and then copy the result into the
+ * context created at the beginning of this function.
+ */
+ datum = ExecEvalExprSwitchContext(exprstate,
+ GetPerTupleExprContext(estate),
+ &isnull);
+ if (isnull)
+ {
+ exprvals[tcnt] = (Datum) 0;
+ exprnulls[tcnt] = true;
+ }
+ else
+ {
+ /* Make sure we copy the data into the context. */
+ Assert(CurrentMemoryContext == expr_context);
+
+ exprvals[tcnt] = datumCopy(datum,
+ stats->attrtype->typbyval,
+ stats->attrtype->typlen);
+ exprnulls[tcnt] = false;
+ }
+
+ tcnt++;
+ }
+
+ /*
+ * Now we can compute the statistics for the expression columns.
+ *
+ * XXX Unlike compute_index_stats we don't need to switch and reset
+ * memory contexts here, because we're only computing stats for a
+ * single expression (and not iterating over many indexes), so we
+ * just do it in expr_context. Note that compute_stats copies the
+ * result into stats->anl_context, so it does not disappear.
+ */
+ if (tcnt > 0)
+ {
+ AttributeOpts *aopt =
+ get_attribute_options(stats->attr->attrelid,
+ stats->attr->attnum);
+
+ stats->exprvals = exprvals;
+ stats->exprnulls = exprnulls;
+ stats->rowstride = 1;
+ stats->compute_stats(stats,
+ expr_fetch_func,
+ tcnt,
+ tcnt);
+
+ /*
+ * If the n_distinct option is specified, it overrides the
+ * above computation.
+ */
+ if (aopt != NULL && aopt->n_distinct != 0.0)
+ stats->stadistinct = aopt->n_distinct;
+ }
+
+ /* And clean up */
+ MemoryContextSwitchTo(expr_context);
+
+ ExecDropSingleTupleTableSlot(slot);
+ FreeExecutorState(estate);
+ MemoryContextResetAndDeleteChildren(expr_context);
+ }
+
+ MemoryContextSwitchTo(old_context);
+ MemoryContextDelete(expr_context);
+}
+
+
+/*
+ * Fetch function for analyzing index expressions.
+ *
+ * We have not bothered to construct index tuples, instead the data is
+ * just in Datum arrays.
+ */
+static Datum
+expr_fetch_func(VacAttrStatsP stats, int rownum, bool *isNull)
+{
+ int i;
+
+ /* exprvals and exprnulls are already offset for proper column */
+ i = rownum * stats->rowstride;
+ *isNull = stats->exprnulls[i];
+ return stats->exprvals[i];
+}
+
+/*
+ * Build analyze data for a list of expressions. As this is not tied
+ * directly to a relation (table or index), we have to fake some of
+ * the data.
+ */
+static AnlExprData *
+build_expr_data(List *exprs)
+{
+ int idx;
+ int nexprs = list_length(exprs);
+ AnlExprData *exprdata;
+ ListCell *lc;
+
+ exprdata = (AnlExprData *) palloc0(nexprs * sizeof(AnlExprData));
+
+ idx = 0;
+ foreach (lc, exprs)
+ {
+ Node *expr = (Node *) lfirst(lc);
+ AnlExprData *thisdata = &exprdata[idx];
+
+ thisdata->expr = expr;
+ thisdata->vacattrstat = (VacAttrStats *) palloc(sizeof(VacAttrStats));
+
+ thisdata->vacattrstat = examine_expression(expr);
+ idx++;
+ }
+
+ return exprdata;
+}
+
+/*
+ * examine_expression -- pre-analysis of a single column
+ *
+ * Determine whether the column is analyzable; if so, create and initialize
+ * a VacAttrStats struct for it. If not, return NULL.
+ */
+static VacAttrStats *
+examine_expression(Node *expr)
+{
+ HeapTuple typtuple;
+ VacAttrStats *stats;
+ int i;
+ bool ok;
+
+ Assert(expr != NULL);
+
+ /*
+ * Create the VacAttrStats struct.
+ */
+ stats = (VacAttrStats *) palloc0(sizeof(VacAttrStats));
+
+ /*
+ * When analyzing an expression, believe the expression tree's type.
+ */
+ stats->attrtypid = exprType(expr);
+ stats->attrtypmod = exprTypmod(expr);
+
+ /*
+ * XXX Do we need to do anything special about the collation, similar
+ * to what examine_attribute does for expression indexes?
+ */
+ stats->attrcollid = exprCollation(expr);
+
+ /*
+ * We don't have any pg_attribute for expressions, so let's fake
+ * something reasonable into attstattarget, which is the only thing
+ * std_typanalyze needs.
+ */
+ stats->attr = (Form_pg_attribute) palloc(ATTRIBUTE_FIXED_PART_SIZE);
+
+ /*
+ * FIXME we should probably get the target from the extended stats
+ * object, or something like that.
+ */
+ stats->attr->attstattarget = default_statistics_target;
+
+ /* initialize some basic fields */
+ stats->attr->attrelid = InvalidOid;
+ stats->attr->attnum = InvalidAttrNumber;
+ stats->attr->atttypid = stats->attrtypid;
+
+ typtuple = SearchSysCacheCopy1(TYPEOID,
+ ObjectIdGetDatum(stats->attrtypid));
+ if (!HeapTupleIsValid(typtuple))
+ elog(ERROR, "cache lookup failed for type %u", stats->attrtypid);
+ stats->attrtype = (Form_pg_type) GETSTRUCT(typtuple);
+ stats->anl_context = CurrentMemoryContext; /* XXX should be using something else? */
+ stats->tupattnum = InvalidAttrNumber;
+
+ /*
+ * The fields describing the stats->stavalues[n] element types default to
+ * the type of the data being analyzed, but the type-specific typanalyze
+ * function can change them if it wants to store something else.
+ */
+ for (i = 0; i < STATISTIC_NUM_SLOTS; i++)
+ {
+ stats->statypid[i] = stats->attrtypid;
+ stats->statyplen[i] = stats->attrtype->typlen;
+ stats->statypbyval[i] = stats->attrtype->typbyval;
+ stats->statypalign[i] = stats->attrtype->typalign;
+ }
+
+ /*
+ * Call the type-specific typanalyze function. If none is specified, use
+ * std_typanalyze().
+ */
+ if (OidIsValid(stats->attrtype->typanalyze))
+ ok = DatumGetBool(OidFunctionCall1(stats->attrtype->typanalyze,
+ PointerGetDatum(stats)));
+ else
+ ok = std_typanalyze(stats);
+
+ if (!ok || stats->compute_stats == NULL || stats->minrows <= 0)
+ {
+ heap_freetuple(typtuple);
+ pfree(stats);
+ return NULL;
+ }
+
+ return stats;
+}
+
+/* form an array of pg_statistic rows (per update_attstats) */
+static Datum
+serialize_expr_stats(AnlExprData *exprdata, int nexprs)
+{
+ int exprno;
+ Oid typOid;
+ Relation sd;
+
+ ArrayBuildState *astate = NULL;
+
+ sd = table_open(StatisticRelationId, RowExclusiveLock);
+
+ /* lookup OID of composite type for pg_statistic */
+ typOid = get_rel_type_id(StatisticRelationId);
+ if (!OidIsValid(typOid))
+ ereport(ERROR,
+ (errcode(ERRCODE_WRONG_OBJECT_TYPE),
+ errmsg("relation \"pg_statistic\" does not have a composite type")));
+
+ for (exprno = 0; exprno < nexprs; exprno++)
+ {
+ int i, k;
+ VacAttrStats *stats = exprdata[exprno].vacattrstat;
+
+ Datum values[Natts_pg_statistic];
+ bool nulls[Natts_pg_statistic];
+ HeapTuple stup;
+
+ if (!stats->stats_valid)
+ {
+ astate = accumArrayResult(astate,
+ (Datum) 0,
+ true,
+ typOid,
+ CurrentMemoryContext);
+ continue;
+ }
+
+ /*
+ * Construct a new pg_statistic tuple
+ */
+ for (i = 0; i < Natts_pg_statistic; ++i)
+ {
+ nulls[i] = false;
+ }
+
+ values[Anum_pg_statistic_starelid - 1] = ObjectIdGetDatum(InvalidOid);
+ values[Anum_pg_statistic_staattnum - 1] = Int16GetDatum(InvalidAttrNumber);
+ values[Anum_pg_statistic_stainherit - 1] = BoolGetDatum(false);
+ values[Anum_pg_statistic_stanullfrac - 1] = Float4GetDatum(stats->stanullfrac);
+ values[Anum_pg_statistic_stawidth - 1] = Int32GetDatum(stats->stawidth);
+ values[Anum_pg_statistic_stadistinct - 1] = Float4GetDatum(stats->stadistinct);
+ i = Anum_pg_statistic_stakind1 - 1;
+ for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+ {
+ values[i++] = Int16GetDatum(stats->stakind[k]); /* stakindN */
+ }
+ i = Anum_pg_statistic_staop1 - 1;
+ for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+ {
+ values[i++] = ObjectIdGetDatum(stats->staop[k]); /* staopN */
+ }
+ i = Anum_pg_statistic_stacoll1 - 1;
+ for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+ {
+ values[i++] = ObjectIdGetDatum(stats->stacoll[k]); /* stacollN */
+ }
+ i = Anum_pg_statistic_stanumbers1 - 1;
+ for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+ {
+ int nnum = stats->numnumbers[k];
+
+ if (nnum > 0)
+ {
+ int n;
+ Datum *numdatums = (Datum *) palloc(nnum * sizeof(Datum));
+ ArrayType *arry;
+
+ for (n = 0; n < nnum; n++)
+ numdatums[n] = Float4GetDatum(stats->stanumbers[k][n]);
+ /* XXX knows more than it should about type float4: */
+ arry = construct_array(numdatums, nnum,
+ FLOAT4OID,
+ sizeof(float4), true, TYPALIGN_INT);
+ values[i++] = PointerGetDatum(arry); /* stanumbersN */
+ }
+ else
+ {
+ nulls[i] = true;
+ values[i++] = (Datum) 0;
+ }
+ }
+ i = Anum_pg_statistic_stavalues1 - 1;
+ for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+ {
+ if (stats->numvalues[k] > 0)
+ {
+ ArrayType *arry;
+
+ arry = construct_array(stats->stavalues[k],
+ stats->numvalues[k],
+ stats->statypid[k],
+ stats->statyplen[k],
+ stats->statypbyval[k],
+ stats->statypalign[k]);
+ values[i++] = PointerGetDatum(arry); /* stavaluesN */
+ }
+ else
+ {
+ nulls[i] = true;
+ values[i++] = (Datum) 0;
+ }
+ }
+
+ stup = heap_form_tuple(RelationGetDescr(sd), values, nulls);
+
+ astate = accumArrayResult(astate,
+ heap_copy_tuple_as_datum(stup, RelationGetDescr(sd)),
+ false,
+ typOid,
+ CurrentMemoryContext);
+ }
+
+ table_close(sd, RowExclusiveLock);
+
+ return makeArrayResult(astate, CurrentMemoryContext);
+}
+
+
+/*
+ * Loads pg_statistic record from expression statistics for expression
+ * identified by the supplied index.
+ */
+HeapTuple
+statext_expressions_load(Oid stxoid, int idx)
+{
+ bool isnull;
+ Datum value;
+ HeapTuple htup;
+ ExpandedArrayHeader *eah;
+ HeapTupleHeader td;
+ HeapTupleData tmptup;
+ HeapTuple tup;
+
+ htup = SearchSysCache1(STATEXTDATASTXOID, ObjectIdGetDatum(stxoid));
+ if (!HeapTupleIsValid(htup))
+ elog(ERROR, "cache lookup failed for statistics object %u", stxoid);
+
+ value = SysCacheGetAttr(STATEXTDATASTXOID, htup,
+ Anum_pg_statistic_ext_data_stxdexpr, &isnull);
+ if (isnull)
+ elog(ERROR,
+ "requested statistics kind \"%c\" is not yet built for statistics object %u",
+ STATS_EXT_DEPENDENCIES, stxoid);
+
+ eah = DatumGetExpandedArray(value);
+
+ deconstruct_expanded_array(eah);
+
+ td = DatumGetHeapTupleHeader(eah->dvalues[idx]);
+
+ /* Build a temporary HeapTuple control structure */
+ tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
+ tmptup.t_data = td;
+
+ tup = heap_copytuple(&tmptup);
+
+ ReleaseSysCache(htup);
+
+ return tup;
+}
+
+/*
+ * Evaluate the expressions, so that we can use the results to build
+ * all the requested statistics types. This matters especially for
+ * expensive expressions, of course.
+ */
+static ExprInfo *
+evaluate_expressions(Relation rel, List *exprs, int numrows, HeapTuple *rows)
+{
+ /* evaluated expressions */
+ ExprInfo *result;
+ char *ptr;
+ Size len;
+
+ int i;
+ int idx;
+ TupleTableSlot *slot;
+ EState *estate;
+ ExprContext *econtext;
+ List *exprstates = NIL;
+ int nexprs = list_length(exprs);
+ ListCell *lc;
+
+ /* allocate everything as a single chunk, so we can free it easily */
+ len = MAXALIGN(sizeof(ExprInfo));
+ len += MAXALIGN(sizeof(Oid) * nexprs); /* types */
+ len += MAXALIGN(sizeof(Oid) * nexprs); /* collations */
+
+ /* values */
+ len += MAXALIGN(sizeof(Datum *) * nexprs);
+ len += nexprs * MAXALIGN(sizeof(Datum) * numrows);
+
+ /* nulls */
+ len += MAXALIGN(sizeof(bool *) * nexprs);
+ len += nexprs * MAXALIGN(sizeof(bool) * numrows);
+
+ ptr = palloc(len);
+
+ /* set the pointers */
+ result = (ExprInfo *) ptr;
+ ptr += MAXALIGN(sizeof(ExprInfo));
+
+ /* types */
+ result->types = (Oid *) ptr;
+ ptr += MAXALIGN(sizeof(Oid) * nexprs);
+
+ /* collations */
+ result->collations = (Oid *) ptr;
+ ptr += MAXALIGN(sizeof(Oid) * nexprs);
+
+ /* values */
+ result->values = (Datum **) ptr;
+ ptr += MAXALIGN(sizeof(Datum *) * nexprs);
+
+ /* nulls */
+ result->nulls = (bool **) ptr;
+ ptr += MAXALIGN(sizeof(bool *) * nexprs);
+
+ for (i = 0; i < nexprs; i++)
+ {
+ result->values[i] = (Datum *) ptr;
+ ptr += MAXALIGN(sizeof(Datum) * numrows);
+
+ result->nulls[i] = (bool *) ptr;
+ ptr += MAXALIGN(sizeof(bool) * numrows);
+ }
+
+ Assert((ptr - (char *) result) == len);
+
+ result->nexprs = list_length(exprs);
+
+ idx = 0;
+ foreach (lc, exprs)
+ {
+ Node *expr = (Node *) lfirst(lc);
+
+ result->types[idx] = exprType(expr);
+ result->collations[idx] = exprCollation(expr);
+
+ idx++;
+ }
+
+ /*
+ * Need an EState for evaluation of index expressions and
+ * partial-index predicates. Create it in the per-index context to be
+ * sure it gets cleaned up at the bottom of the loop.
+ */
+ estate = CreateExecutorState();
+ econtext = GetPerTupleExprContext(estate);
+
+ /* Need a slot to hold the current heap tuple, too */
+ slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
+
+ /* Arrange for econtext's scan tuple to be the tuple under test */
+ econtext->ecxt_scantuple = slot;
+
+ /* Set up expression evaluation state */
+ exprstates = ExecPrepareExprList(exprs, estate);
+
+ for (i = 0; i < numrows; i++)
+ {
+ /*
+ * Reset the per-tuple context each time, to reclaim any cruft
+ * left behind by evaluating the predicate or index expressions.
+ */
+ ResetExprContext(econtext);
+
+ /* Set up for predicate or expression evaluation */
+ ExecStoreHeapTuple(rows[i], slot, false);
+
+ idx = 0;
+ foreach (lc, exprstates)
+ {
+ Datum datum;
+ bool isnull;
+ ExprState *exprstate = (ExprState *) lfirst(lc);
+
+ /*
+ * FIXME this probably leaks memory. Maybe we should use
+ * ExecEvalExprSwitchContext but then we need to copy the
+ * result somewhere else.
+ */
+ datum = ExecEvalExpr(exprstate,
+ GetPerTupleExprContext(estate),
+ &isnull);
+ if (isnull)
+ {
+ result->values[idx][i] = (Datum) 0;
+ result->nulls[idx][i] = true;
+ }
+ else
+ {
+ result->values[idx][i] = (Datum) datum;
+ result->nulls[idx][i] = false;
+ }
+
+ idx++;
+ }
+ }
+
+ ExecDropSingleTupleTableSlot(slot);
+ FreeExecutorState(estate);
+
+ return result;
+}
+
+/*
+ * add_expressions_to_attributes
+ * add expressions as attributes with high attnums
+ *
+ * Treat the expressions as attributes with attnums above the regular
+ * attnum range. This will allow us to handle everything in the same
+ * way, and identify expressions in the dependencies.
+ *
+ * XXX This always creates a copy of the bitmap. We might optimize this
+ * by only creating the copy with (nexprs > 0) but then we'd have to track
+ * this in order to free it (if we want to). Does not seem worth it.
+ */
+Bitmapset *
+add_expressions_to_attributes(Bitmapset *attrs, int nexprs)
+{
+ int i;
+
+ /*
+ * Copy the bitmapset and add fake attnums representing expressions,
+ * starting above MaxHeapAttributeNumber.
+ */
+ attrs = bms_copy(attrs);
+
+ /* start with (MaxHeapAttributeNumber + 1) */
+ for (i = 0; i < nexprs; i++)
+ {
+ Assert(EXPRESSION_ATTNUM(i) > MaxHeapAttributeNumber);
+
+ attrs = bms_add_member(attrs, EXPRESSION_ATTNUM(i));
+ }
+
+ return attrs;
+}
diff --git a/src/backend/statistics/mcv.c b/src/backend/statistics/mcv.c
index abbc1f1ba8..0c27ee395e 100644
--- a/src/backend/statistics/mcv.c
+++ b/src/backend/statistics/mcv.c
@@ -74,7 +74,8 @@
((ndims) * sizeof(DimensionInfo)) + \
((nitems) * ITEM_SIZE(ndims)))
-static MultiSortSupport build_mss(VacAttrStats **stats, int numattrs);
+static MultiSortSupport build_mss(VacAttrStats **stats, int numattrs,
+ ExprInfo *exprs);
static SortItem *build_distinct_groups(int numrows, SortItem *items,
MultiSortSupport mss, int *ndistinct);
@@ -181,8 +182,9 @@ get_mincount_for_mcv_list(int samplerows, double totalrows)
*
*/
MCVList *
-statext_mcv_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
- VacAttrStats **stats, double totalrows, int stattarget)
+statext_mcv_build(int numrows, HeapTuple *rows, ExprInfo *exprs,
+ Bitmapset *attrs, VacAttrStats **stats,
+ double totalrows, int stattarget)
{
int i,
numattrs,
@@ -195,14 +197,23 @@ statext_mcv_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
MCVList *mcvlist = NULL;
MultiSortSupport mss;
- attnums = build_attnums_array(attrs, &numattrs);
-
/* comparator for all the columns */
- mss = build_mss(stats, numattrs);
+ mss = build_mss(stats, bms_num_members(attrs), exprs);
+
+ /*
+ * treat expressions as special attributes with high attnums
+ *
+ * XXX We do this after build_mss, because that expects the bitmapset
+ * to only contain simple attributes (with a matching VacAttrStats)
+ */
+ attrs = add_expressions_to_attributes(attrs, exprs->nexprs);
+
+ /* now build the array, with the special expression attnums */
+ attnums = build_attnums_array(attrs, &numattrs);
/* sort the rows */
- items = build_sorted_items(numrows, &nitems, rows, stats[0]->tupDesc,
- mss, numattrs, attnums);
+ items = build_sorted_items(numrows, &nitems, rows, exprs,
+ stats[0]->tupDesc, mss, numattrs, attnums);
if (!items)
return NULL;
@@ -338,6 +349,7 @@ statext_mcv_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
pfree(items);
pfree(groups);
+ pfree(attrs);
return mcvlist;
}
@@ -347,12 +359,12 @@ statext_mcv_build(int numrows, HeapTuple *rows, Bitmapset *attrs,
* build MultiSortSupport for the attributes passed in attrs
*/
static MultiSortSupport
-build_mss(VacAttrStats **stats, int numattrs)
+build_mss(VacAttrStats **stats, int numattrs, ExprInfo *exprs)
{
int i;
/* Sort by multiple columns (using array of SortSupport) */
- MultiSortSupport mss = multi_sort_init(numattrs);
+ MultiSortSupport mss = multi_sort_init(numattrs + exprs->nexprs);
/* prepare the sort functions for all the attributes */
for (i = 0; i < numattrs; i++)
@@ -368,6 +380,20 @@ build_mss(VacAttrStats **stats, int numattrs)
multi_sort_add_dimension(mss, i, type->lt_opr, colstat->attrcollid);
}
+ /* prepare the sort functions for all the expressions */
+ for (i = 0; i < exprs->nexprs; i++)
+ {
+ TypeCacheEntry *type;
+
+ type = lookup_type_cache(exprs->types[i], TYPECACHE_LT_OPR);
+ if (type->lt_opr == InvalidOid) /* shouldn't happen */
+ elog(ERROR, "cache lookup failed for ordering operator for type %u",
+ exprs->types[i]);
+
+ multi_sort_add_dimension(mss, numattrs + i, type->lt_opr,
+ exprs->collations[i]);
+ }
+
return mss;
}
@@ -570,7 +596,7 @@ statext_mcv_load(Oid mvoid)
if (isnull)
elog(ERROR,
- "requested statistic kind \"%c\" is not yet built for statistics object %u",
+ "requested statistics kind \"%c\" is not yet built for statistics object %u",
STATS_EXT_DEPENDENCIES, mvoid);
result = statext_mcv_deserialize(DatumGetByteaP(mcvlist));
@@ -1541,10 +1567,14 @@ pg_mcv_list_send(PG_FUNCTION_ARGS)
* the size to ~1/8. It would also allow us to combine bitmaps simply using
* & and |, which should be faster than min/max. The bitmaps are fairly
* small, though (thanks to the cap on the MCV list size).
+ *
+ * XXX There's a lot of code duplication between branches for simple columns
+ * and complex expressions. We should refactor it somehow.
*/
static bool *
mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
- Bitmapset *keys, MCVList *mcvlist, bool is_or)
+ Bitmapset *keys, List *exprs,
+ MCVList *mcvlist, bool is_or)
{
int i;
ListCell *l;
@@ -1584,8 +1614,10 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
/* valid only after examine_clause_args returns true */
Var *var;
+ Node *clause_expr;
Const *cst;
bool varonleft;
+ bool expronleft;
fmgr_info(get_opcode(expr->opno), &opproc);
@@ -1654,6 +1686,89 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
matches[i] = RESULT_MERGE(matches[i], is_or, match);
}
}
+ /* extract the expr and const from the expression */
+ else if (examine_clause_args2(expr->args, &clause_expr, &cst, &expronleft))
+ {
+ ListCell *lc;
+ int idx;
+ Oid collid = exprCollation(clause_expr);
+
+ /* match the attribute to a dimension of the statistic */
+ idx = bms_num_members(keys);
+
+ foreach(lc, exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc);
+
+ if (equal(clause_expr, stat_expr))
+ break;
+
+ idx++;
+ }
+
+ /* index should be valid */
+ Assert((idx >= 0) &&
+ (idx < bms_num_members(keys) + list_length(exprs)));
+
+ /*
+ * Walk through the MCV items and evaluate the current clause.
+ * We can skip items that were already ruled out, and
+ * terminate if there are no remaining MCV items that might
+ * possibly match.
+ */
+ for (i = 0; i < mcvlist->nitems; i++)
+ {
+ bool match = true;
+ MCVItem *item = &mcvlist->items[i];
+
+ /*
+ * When the MCV item or the Const value is NULL we can
+ * treat this as a mismatch. We must not call the operator
+ * because of strictness.
+ */
+ if (item->isnull[idx] || cst->constisnull)
+ {
+ matches[i] = RESULT_MERGE(matches[i], is_or, false);
+ continue;
+ }
+
+ /*
+ * Skip MCV items that can't change result in the bitmap.
+ * Once the value gets false for AND-lists, or true for
+ * OR-lists, we don't need to look at more clauses.
+ */
+ if (RESULT_IS_FINAL(matches[i], is_or))
+ continue;
+
+ /*
+ * First check whether the constant is below the lower
+ * boundary (in that case we can skip the bucket, because
+ * there's no overlap).
+ *
+ * We don't store collations used to build the statistics,
+ * but we can use the collation for the attribute itself,
+ * as stored in varcollid. We do reset the statistics
+ * after a type change (including collation change), so
+ * this is OK. We may need to relax this after allowing
+ * extended statistics on expressions.
+ */
+ if (expronleft)
+ match = DatumGetBool(FunctionCall2Coll(&opproc,
+ collid,
+ item->values[idx],
+ cst->constvalue));
+ else
+ match = DatumGetBool(FunctionCall2Coll(&opproc,
+ collid,
+ cst->constvalue,
+ item->values[idx]));
+
+ /* update the match bitmap with the result */
+ matches[i] = RESULT_MERGE(matches[i], is_or, match);
+ }
+ }
+ else
+ elog(ERROR, "incompatible clause");
}
else if (IsA(clause, ScalarArrayOpExpr))
{
@@ -1662,8 +1777,10 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
/* valid only after examine_clause_args returns true */
Var *var;
+ Node *clause_expr;
Const *cst;
bool varonleft;
+ bool expronleft;
fmgr_info(get_opcode(expr->opno), &opproc);
@@ -1761,14 +1878,155 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
matches[i] = RESULT_MERGE(matches[i], is_or, match);
}
}
+ /* extract the expr and const from the expression */
+ else if (examine_clause_args2(expr->args, &clause_expr, &cst, &expronleft))
+ {
+ ListCell *lc;
+ int idx;
+
+ ArrayType *arrayval;
+ int16 elmlen;
+ bool elmbyval;
+ char elmalign;
+ int num_elems;
+ Datum *elem_values;
+ bool *elem_nulls;
+ Oid collid = exprCollation(clause_expr);
+
+ /* ScalarArrayOpExpr has the Var always on the left */
+ Assert(expronleft);
+
+ if (!cst->constisnull)
+ {
+ arrayval = DatumGetArrayTypeP(cst->constvalue);
+ get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
+ &elmlen, &elmbyval, &elmalign);
+ deconstruct_array(arrayval,
+ ARR_ELEMTYPE(arrayval),
+ elmlen, elmbyval, elmalign,
+ &elem_values, &elem_nulls, &num_elems);
+ }
+
+ /* match the attribute to a dimension of the statistic */
+ idx = bms_num_members(keys);
+
+ foreach(lc, exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc);
+
+ if (equal(clause_expr, stat_expr))
+ break;
+
+ idx++;
+ }
+
+ /* index should be valid */
+ Assert((idx >= 0) &&
+ (idx < bms_num_members(keys) + list_length(exprs)));
+
+ /*
+ * Walk through the MCV items and evaluate the current clause.
+ * We can skip items that were already ruled out, and
+ * terminate if there are no remaining MCV items that might
+ * possibly match.
+ */
+ for (i = 0; i < mcvlist->nitems; i++)
+ {
+ int j;
+ bool match = (expr->useOr ? false : true);
+ MCVItem *item = &mcvlist->items[i];
+
+ /*
+ * When the MCV item or the Const value is NULL we can
+ * treat this as a mismatch. We must not call the operator
+ * because of strictness.
+ */
+ if (item->isnull[idx] || cst->constisnull)
+ {
+ matches[i] = RESULT_MERGE(matches[i], is_or, false);
+ continue;
+ }
+
+ /*
+ * Skip MCV items that can't change result in the bitmap.
+ * Once the value gets false for AND-lists, or true for
+ * OR-lists, we don't need to look at more clauses.
+ */
+ if (RESULT_IS_FINAL(matches[i], is_or))
+ continue;
+
+ for (j = 0; j < num_elems; j++)
+ {
+ Datum elem_value = elem_values[j];
+ bool elem_isnull = elem_nulls[j];
+ bool elem_match;
+
+ /* NULL values always evaluate as not matching. */
+ if (elem_isnull)
+ {
+ match = RESULT_MERGE(match, expr->useOr, false);
+ continue;
+ }
+
+ /*
+ * Stop evaluating the array elements once we reach
+ * match value that can't change - ALL() is the same
+ * as AND-list, ANY() is the same as OR-list.
+ */
+ if (RESULT_IS_FINAL(match, expr->useOr))
+ break;
+
+ elem_match = DatumGetBool(FunctionCall2Coll(&opproc,
+ collid,
+ item->values[idx],
+ elem_value));
+
+ match = RESULT_MERGE(match, expr->useOr, elem_match);
+ }
+
+ /* update the match bitmap with the result */
+ matches[i] = RESULT_MERGE(matches[i], is_or, match);
+ }
+ }
+ else
+ elog(ERROR, "incompatible clause");
}
else if (IsA(clause, NullTest))
{
NullTest *expr = (NullTest *) clause;
- Var *var = (Var *) (expr->arg);
+ Node *clause_expr = (Node *) (expr->arg);
/* match the attribute to a dimension of the statistic */
- int idx = bms_member_index(keys, var->varattno);
+ int idx = -1;
+
+ if (IsA(clause_expr, Var))
+ {
+ /* simple Var, so just lookup using varattno */
+ Var *var = (Var *) clause_expr;
+
+ idx = bms_member_index(keys, var->varattno);
+ }
+ else
+ {
+ ListCell *lc;
+
+ /* expressions are after the simple columns */
+ idx = bms_num_members(keys);
+
+ /* expression - lookup in stats expressions */
+ foreach(lc, exprs)
+ {
+ Node *stat_expr = (Node *) lfirst(lc);
+
+ if (equal(clause_expr, stat_expr))
+ break;
+
+ idx++;
+ }
+ }
+
+ /* index should be valid */
+ Assert((idx >= 0) && (idx < bms_num_members(keys) + list_length(exprs)));
/*
* Walk through the MCV items and evaluate the current clause. We
@@ -1811,7 +2069,7 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
Assert(list_length(bool_clauses) >= 2);
/* build the match bitmap for the OR-clauses */
- bool_matches = mcv_get_match_bitmap(root, bool_clauses, keys,
+ bool_matches = mcv_get_match_bitmap(root, bool_clauses, keys, exprs,
mcvlist, is_orclause(clause));
/*
@@ -1839,7 +2097,7 @@ mcv_get_match_bitmap(PlannerInfo *root, List *clauses,
Assert(list_length(not_args) == 1);
/* build the match bitmap for the NOT-clause */
- not_matches = mcv_get_match_bitmap(root, not_args, keys,
+ not_matches = mcv_get_match_bitmap(root, not_args, keys, exprs,
mcvlist, false);
/*
@@ -1982,7 +2240,8 @@ mcv_clauselist_selectivity(PlannerInfo *root, StatisticExtInfo *stat,
mcv = statext_mcv_load(stat->statOid);
/* build a match bitmap for the clauses */
- matches = mcv_get_match_bitmap(root, clauses, stat->keys, mcv, false);
+ matches = mcv_get_match_bitmap(root, clauses, stat->keys, stat->exprs,
+ mcv, false);
/* sum frequencies for all the matching MCV items */
*basesel = 0.0;
@@ -2056,7 +2315,7 @@ mcv_clause_selectivity_or(PlannerInfo *root, StatisticExtInfo *stat,
/* build the match bitmap for the new clause */
new_matches = mcv_get_match_bitmap(root, list_make1(clause), stat->keys,
- mcv, false);
+ stat->exprs, mcv, false);
/*
* Sum the frequencies for all the MCV items matching this clause and also
diff --git a/src/backend/statistics/mvdistinct.c b/src/backend/statistics/mvdistinct.c
index 9ef21debb6..55d3fa0e1f 100644
--- a/src/backend/statistics/mvdistinct.c
+++ b/src/backend/statistics/mvdistinct.c
@@ -37,7 +37,8 @@
#include "utils/typcache.h"
static double ndistinct_for_combination(double totalrows, int numrows,
- HeapTuple *rows, VacAttrStats **stats,
+ HeapTuple *rows, ExprInfo *exprs,
+ int nattrs, VacAttrStats **stats,
int k, int *combination);
static double estimate_ndistinct(double totalrows, int numrows, int d, int f1);
static int n_choose_k(int n, int k);
@@ -81,16 +82,21 @@ static void generate_combinations(CombinationGenerator *state);
*
* This computes the ndistinct estimate using the same estimator used
* in analyze.c and then computes the coefficient.
+ *
+ * To handle expressions easily, we treat them as special attributes with
+ * attnums above MaxHeapAttributeNumber, and we assume the expressions are
+ * placed after all simple attributes.
*/
MVNDistinct *
statext_ndistinct_build(double totalrows, int numrows, HeapTuple *rows,
- Bitmapset *attrs, VacAttrStats **stats)
+ ExprInfo *exprs, Bitmapset *attrs,
+ VacAttrStats **stats)
{
MVNDistinct *result;
int k;
int itemcnt;
int numattrs = bms_num_members(attrs);
- int numcombs = num_combinations(numattrs);
+ int numcombs = num_combinations(numattrs + exprs->nexprs);
result = palloc(offsetof(MVNDistinct, items) +
numcombs * sizeof(MVNDistinctItem));
@@ -98,14 +104,20 @@ statext_ndistinct_build(double totalrows, int numrows, HeapTuple *rows,
result->type = STATS_NDISTINCT_TYPE_BASIC;
result->nitems = numcombs;
+ /* treat expressions as special attributes with high attnums */
+ attrs = add_expressions_to_attributes(attrs, exprs->nexprs);
+
+ /* make sure there were no clashes */
+ Assert(bms_num_members(attrs) == numattrs + exprs->nexprs);
+
itemcnt = 0;
- for (k = 2; k <= numattrs; k++)
+ for (k = 2; k <= bms_num_members(attrs); k++)
{
int *combination;
CombinationGenerator *generator;
/* generate combinations of K out of N elements */
- generator = generator_init(numattrs, k);
+ generator = generator_init(bms_num_members(attrs), k);
while ((combination = generator_next(generator)))
{
@@ -114,10 +126,32 @@ statext_ndistinct_build(double totalrows, int numrows, HeapTuple *rows,
item->attrs = NULL;
for (j = 0; j < k; j++)
- item->attrs = bms_add_member(item->attrs,
- stats[combination[j]]->attr->attnum);
+ {
+ AttrNumber attnum = InvalidAttrNumber;
+
+ /*
+ * The simple attributes are before expressions, so have
+ * indexes below numattrs.
+ * */
+ if (combination[j] < numattrs)
+ attnum = stats[combination[j]]->attr->attnum;
+ else
+ {
+ /* make sure the expression index is valid */
+ Assert((combination[j] - numattrs) >= 0);
+ Assert((combination[j] - numattrs) < exprs->nexprs);
+
+ attnum = EXPRESSION_ATTNUM(combination[j] - numattrs);
+ }
+
+ Assert(attnum != InvalidAttrNumber);
+
+ item->attrs = bms_add_member(item->attrs, attnum);
+ }
+
item->ndistinct =
ndistinct_for_combination(totalrows, numrows, rows,
+ exprs, numattrs,
stats, k, combination);
itemcnt++;
@@ -153,7 +187,7 @@ statext_ndistinct_load(Oid mvoid)
Anum_pg_statistic_ext_data_stxdndistinct, &isnull);
if (isnull)
elog(ERROR,
- "requested statistic kind \"%c\" is not yet built for statistics object %u",
+ "requested statistics kind \"%c\" is not yet built for statistics object %u",
STATS_EXT_NDISTINCT, mvoid);
result = statext_ndistinct_deserialize(DatumGetByteaPP(ndist));
@@ -428,6 +462,7 @@ pg_ndistinct_send(PG_FUNCTION_ARGS)
*/
static double
ndistinct_for_combination(double totalrows, int numrows, HeapTuple *rows,
+ ExprInfo *exprs, int nattrs,
VacAttrStats **stats, int k, int *combination)
{
int i,
@@ -467,25 +502,57 @@ ndistinct_for_combination(double totalrows, int numrows, HeapTuple *rows,
*/
for (i = 0; i < k; i++)
{
- VacAttrStats *colstat = stats[combination[i]];
+ Oid typid;
TypeCacheEntry *type;
+ AttrNumber attnum = InvalidAttrNumber;
+ TupleDesc tdesc = NULL;
+ Oid collid = InvalidOid;
+
+ if (combination[i] < nattrs)
+ {
+ VacAttrStats *colstat = stats[combination[i]];
+ typid = colstat->attrtypid;
+ attnum = colstat->attr->attnum;
+ collid = colstat->attrcollid;
+ tdesc = colstat->tupDesc;
+ }
+ else
+ {
+ typid = exprs->types[combination[i] - nattrs];
+ collid = exprs->collations[combination[i] - nattrs];
+ }
- type = lookup_type_cache(colstat->attrtypid, TYPECACHE_LT_OPR);
+ type = lookup_type_cache(typid, TYPECACHE_LT_OPR);
if (type->lt_opr == InvalidOid) /* shouldn't happen */
elog(ERROR, "cache lookup failed for ordering operator for type %u",
- colstat->attrtypid);
+ typid);
/* prepare the sort function for this dimension */
- multi_sort_add_dimension(mss, i, type->lt_opr, colstat->attrcollid);
+ multi_sort_add_dimension(mss, i, type->lt_opr, collid);
/* accumulate all the data for this dimension into the arrays */
for (j = 0; j < numrows; j++)
{
- items[j].values[i] =
- heap_getattr(rows[j],
- colstat->attr->attnum,
- colstat->tupDesc,
- &items[j].isnull[i]);
+ /*
+ * The first nattrs indexes identify simple attributes, higher
+ * indexes are expressions.
+ */
+ if (combination[i] < nattrs)
+ items[j].values[i] =
+ heap_getattr(rows[j],
+ attnum,
+ tdesc,
+ &items[j].isnull[i]);
+ else
+ {
+ int idx = (combination[i] - nattrs);
+
+ /* make sure the expression index is valid */
+ Assert((idx >= 0) && (idx < exprs->nexprs));
+
+ items[j].values[i] = exprs->values[idx][j];
+ items[j].isnull[i] = exprs->nulls[idx][j];
+ }
}
}
diff --git a/src/backend/tcop/utility.c b/src/backend/tcop/utility.c
index 1d81071c35..eb0c030025 100644
--- a/src/backend/tcop/utility.c
+++ b/src/backend/tcop/utility.c
@@ -1796,7 +1796,28 @@ ProcessUtilitySlow(ParseState *pstate,
break;
case T_CreateStatsStmt:
- address = CreateStatistics((CreateStatsStmt *) parsetree);
+ {
+ Oid relid;
+ CreateStatsStmt *stmt = (CreateStatsStmt *) parsetree;
+ RangeVar *rel = (RangeVar *) linitial(stmt->relations);
+
+ /*
+ * CREATE STATISTICS will influence future execution plans
+ * but does not interfere with currently executing plans.
+ * So it should be enough to take ShareUpdateExclusiveLock
+ * on relation, conflicting with ANALYZE and other DDL that
+ * sets statistical information, but not with normal queries.
+ *
+ * XXX RangeVarCallbackOwnsRelation not needed needed here,
+ * to keep the same behavior as before.
+ */
+ relid = RangeVarGetRelid(rel, ShareUpdateExclusiveLock, false);
+
+ /* Run parse analysis ... */
+ stmt = transformStatsStmt(relid, stmt, queryString);
+
+ address = CreateStatistics(stmt);
+ }
break;
case T_AlterStatsStmt:
diff --git a/src/backend/utils/adt/ruleutils.c b/src/backend/utils/adt/ruleutils.c
index 8a1fbda572..7d08d752a1 100644
--- a/src/backend/utils/adt/ruleutils.c
+++ b/src/backend/utils/adt/ruleutils.c
@@ -336,7 +336,8 @@ static char *pg_get_indexdef_worker(Oid indexrelid, int colno,
bool attrsOnly, bool keysOnly,
bool showTblSpc, bool inherits,
int prettyFlags, bool missing_ok);
-static char *pg_get_statisticsobj_worker(Oid statextid, bool missing_ok);
+static char *pg_get_statisticsobj_worker(Oid statextid, bool columns_only,
+ bool missing_ok);
static char *pg_get_partkeydef_worker(Oid relid, int prettyFlags,
bool attrsOnly, bool missing_ok);
static char *pg_get_constraintdef_worker(Oid constraintId, bool fullCommand,
@@ -1507,7 +1508,26 @@ pg_get_statisticsobjdef(PG_FUNCTION_ARGS)
Oid statextid = PG_GETARG_OID(0);
char *res;
- res = pg_get_statisticsobj_worker(statextid, true);
+ res = pg_get_statisticsobj_worker(statextid, false, true);
+
+ if (res == NULL)
+ PG_RETURN_NULL();
+
+ PG_RETURN_TEXT_P(string_to_text(res));
+}
+
+
+/*
+ * pg_get_statisticsobjdef_columns
+ * Get columns and expressions for an extended statistics object
+ */
+Datum
+pg_get_statisticsobjdef_columns(PG_FUNCTION_ARGS)
+{
+ Oid statextid = PG_GETARG_OID(0);
+ char *res;
+
+ res = pg_get_statisticsobj_worker(statextid, true, true);
if (res == NULL)
PG_RETURN_NULL();
@@ -1519,7 +1539,7 @@ pg_get_statisticsobjdef(PG_FUNCTION_ARGS)
* Internal workhorse to decompile an extended statistics object.
*/
static char *
-pg_get_statisticsobj_worker(Oid statextid, bool missing_ok)
+pg_get_statisticsobj_worker(Oid statextid, bool columns_only, bool missing_ok)
{
Form_pg_statistic_ext statextrec;
HeapTuple statexttup;
@@ -1534,6 +1554,11 @@ pg_get_statisticsobj_worker(Oid statextid, bool missing_ok)
bool dependencies_enabled;
bool mcv_enabled;
int i;
+ List *context;
+ ListCell *lc;
+ List *exprs = NIL;
+ bool has_exprs;
+ int ncolumns;
statexttup = SearchSysCache1(STATEXTOID, ObjectIdGetDatum(statextid));
@@ -1544,75 +1569,112 @@ pg_get_statisticsobj_worker(Oid statextid, bool missing_ok)
elog(ERROR, "cache lookup failed for statistics object %u", statextid);
}
- statextrec = (Form_pg_statistic_ext) GETSTRUCT(statexttup);
-
- initStringInfo(&buf);
+ /* has the statistics expressions? */
+ has_exprs = !heap_attisnull(statexttup, Anum_pg_statistic_ext_stxexprs, NULL);
- nsp = get_namespace_name(statextrec->stxnamespace);
- appendStringInfo(&buf, "CREATE STATISTICS %s",
- quote_qualified_identifier(nsp,
- NameStr(statextrec->stxname)));
+ statextrec = (Form_pg_statistic_ext) GETSTRUCT(statexttup);
/*
- * Decode the stxkind column so that we know which stats types to print.
+ * Get the statistics expressions, if any. (NOTE: we do not use the
+ * relcache versions of the expressions and predicate, because we want
+ * to display non-const-folded expressions.)
*/
- datum = SysCacheGetAttr(STATEXTOID, statexttup,
- Anum_pg_statistic_ext_stxkind, &isnull);
- Assert(!isnull);
- arr = DatumGetArrayTypeP(datum);
- if (ARR_NDIM(arr) != 1 ||
- ARR_HASNULL(arr) ||
- ARR_ELEMTYPE(arr) != CHAROID)
- elog(ERROR, "stxkind is not a 1-D char array");
- enabled = (char *) ARR_DATA_PTR(arr);
-
- ndistinct_enabled = false;
- dependencies_enabled = false;
- mcv_enabled = false;
-
- for (i = 0; i < ARR_DIMS(arr)[0]; i++)
+ if (has_exprs)
{
- if (enabled[i] == STATS_EXT_NDISTINCT)
- ndistinct_enabled = true;
- if (enabled[i] == STATS_EXT_DEPENDENCIES)
- dependencies_enabled = true;
- if (enabled[i] == STATS_EXT_MCV)
- mcv_enabled = true;
+ Datum exprsDatum;
+ bool isnull;
+ char *exprsString;
+
+ exprsDatum = SysCacheGetAttr(STATEXTOID, statexttup,
+ Anum_pg_statistic_ext_stxexprs, &isnull);
+ Assert(!isnull);
+ exprsString = TextDatumGetCString(exprsDatum);
+ exprs = (List *) stringToNode(exprsString);
+ pfree(exprsString);
}
+ else
+ exprs = NIL;
- /*
- * If any option is disabled, then we'll need to append the types clause
- * to show which options are enabled. We omit the types clause on purpose
- * when all options are enabled, so a pg_dump/pg_restore will create all
- * statistics types on a newer postgres version, if the statistics had all
- * options enabled on the original version.
- */
- if (!ndistinct_enabled || !dependencies_enabled || !mcv_enabled)
+ /* count the number of columns (attributes and expressions) */
+ ncolumns = statextrec->stxkeys.dim1 + list_length(exprs);
+
+ initStringInfo(&buf);
+
+ if (!columns_only)
{
- bool gotone = false;
+ nsp = get_namespace_name(statextrec->stxnamespace);
+ appendStringInfo(&buf, "CREATE STATISTICS %s",
+ quote_qualified_identifier(nsp,
+ NameStr(statextrec->stxname)));
- appendStringInfoString(&buf, " (");
+ /*
+ * Decode the stxkind column so that we know which stats types to print.
+ */
+ datum = SysCacheGetAttr(STATEXTOID, statexttup,
+ Anum_pg_statistic_ext_stxkind, &isnull);
+ Assert(!isnull);
+ arr = DatumGetArrayTypeP(datum);
+ if (ARR_NDIM(arr) != 1 ||
+ ARR_HASNULL(arr) ||
+ ARR_ELEMTYPE(arr) != CHAROID)
+ elog(ERROR, "stxkind is not a 1-D char array");
+ enabled = (char *) ARR_DATA_PTR(arr);
+
+ ndistinct_enabled = false;
+ dependencies_enabled = false;
+ mcv_enabled = false;
- if (ndistinct_enabled)
+ for (i = 0; i < ARR_DIMS(arr)[0]; i++)
{
- appendStringInfoString(&buf, "ndistinct");
- gotone = true;
+ if (enabled[i] == STATS_EXT_NDISTINCT)
+ ndistinct_enabled = true;
+ else if (enabled[i] == STATS_EXT_DEPENDENCIES)
+ dependencies_enabled = true;
+ else if (enabled[i] == STATS_EXT_MCV)
+ mcv_enabled = true;
+
+ /* ignore STATS_EXT_EXPRESSIONS (it's built automatically) */
}
- if (dependencies_enabled)
+ /*
+ * If any option is disabled, then we'll need to append the types clause
+ * to show which options are enabled. We omit the types clause on purpose
+ * when all options are enabled, so a pg_dump/pg_restore will create all
+ * statistics types on a newer postgres version, if the statistics had all
+ * options enabled on the original version.
+ *
+ * But if the statistics is defined on just a single column, it has to be
+ * an expression statistics. In that case we don't need to specify kinds.
+ */
+ if ((!ndistinct_enabled || !dependencies_enabled || !mcv_enabled) &&
+ (ncolumns > 1))
{
- appendStringInfo(&buf, "%sdependencies", gotone ? ", " : "");
- gotone = true;
- }
+ bool gotone = false;
- if (mcv_enabled)
- appendStringInfo(&buf, "%smcv", gotone ? ", " : "");
+ appendStringInfoString(&buf, " (");
- appendStringInfoChar(&buf, ')');
- }
+ if (ndistinct_enabled)
+ {
+ appendStringInfoString(&buf, "ndistinct");
+ gotone = true;
+ }
+
+ if (dependencies_enabled)
+ {
+ appendStringInfo(&buf, "%sdependencies", gotone ? ", " : "");
+ gotone = true;
+ }
- appendStringInfoString(&buf, " ON ");
+ if (mcv_enabled)
+ appendStringInfo(&buf, "%smcv", gotone ? ", " : "");
+ appendStringInfoChar(&buf, ')');
+ }
+
+ appendStringInfoString(&buf, " ON ");
+ }
+
+ /* decode simple column references */
for (colno = 0; colno < statextrec->stxkeys.dim1; colno++)
{
AttrNumber attnum = statextrec->stxkeys.values[colno];
@@ -1626,14 +1688,109 @@ pg_get_statisticsobj_worker(Oid statextid, bool missing_ok)
appendStringInfoString(&buf, quote_identifier(attname));
}
- appendStringInfo(&buf, " FROM %s",
- generate_relation_name(statextrec->stxrelid, NIL));
+ context = deparse_context_for(get_relation_name(statextrec->stxrelid),
+ statextrec->stxrelid);
+
+ foreach (lc, exprs)
+ {
+ Node *expr = (Node *) lfirst(lc);
+ char *str;
+ int prettyFlags = PRETTYFLAG_INDENT;
+
+ str = deparse_expression_pretty(expr, context, false, false,
+ prettyFlags, 0);
+
+ if (colno > 0)
+ appendStringInfoString(&buf, ", ");
+
+ /* Need parens if it's not a bare function call */
+ if (looks_like_function(expr))
+ appendStringInfoString(&buf, str);
+ else
+ appendStringInfo(&buf, "(%s)", str);
+
+ colno++;
+ }
+
+ if (!columns_only)
+ appendStringInfo(&buf, " FROM %s",
+ generate_relation_name(statextrec->stxrelid, NIL));
ReleaseSysCache(statexttup);
return buf.data;
}
+/*
+ * Generate text array of expressions for statistics object.
+ */
+Datum
+pg_get_statisticsobjdef_expressions(PG_FUNCTION_ARGS)
+{
+ Oid statextid = PG_GETARG_OID(0);
+ Form_pg_statistic_ext statextrec;
+ HeapTuple statexttup;
+ Datum datum;
+ bool isnull;
+ List *context;
+ ListCell *lc;
+ List *exprs = NIL;
+ bool has_exprs;
+ char *tmp;
+ ArrayBuildState *astate = NULL;
+
+ statexttup = SearchSysCache1(STATEXTOID, ObjectIdGetDatum(statextid));
+
+ if (!HeapTupleIsValid(statexttup))
+ elog(ERROR, "cache lookup failed for statistics object %u", statextid);
+
+ /* has the statistics expressions? */
+ has_exprs = !heap_attisnull(statexttup, Anum_pg_statistic_ext_stxexprs, NULL);
+
+ /* no expressions? we're done */
+ if (!has_exprs)
+ {
+ ReleaseSysCache(statexttup);
+ PG_RETURN_NULL();
+ }
+
+ statextrec = (Form_pg_statistic_ext) GETSTRUCT(statexttup);
+
+ /*
+ * Get the statistics expressions, and deparse them into text values.
+ */
+ datum = SysCacheGetAttr(STATEXTOID, statexttup,
+ Anum_pg_statistic_ext_stxexprs, &isnull);
+
+ Assert(!isnull);
+ tmp = TextDatumGetCString(datum);
+ exprs = (List *) stringToNode(tmp);
+ pfree(tmp);
+
+ context = deparse_context_for(get_relation_name(statextrec->stxrelid),
+ statextrec->stxrelid);
+
+ foreach (lc, exprs)
+ {
+ Node *expr = (Node *) lfirst(lc);
+ char *str;
+ int prettyFlags = PRETTYFLAG_INDENT;
+
+ str = deparse_expression_pretty(expr, context, false, false,
+ prettyFlags, 0);
+
+ astate = accumArrayResult(astate,
+ PointerGetDatum(cstring_to_text(str)),
+ false,
+ TEXTOID,
+ CurrentMemoryContext);
+ }
+
+ ReleaseSysCache(statexttup);
+
+ PG_RETURN_DATUM(makeArrayResult(astate, CurrentMemoryContext));
+}
+
/*
* pg_get_partkeydef
*
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index d5e61664bc..bfe3dfb93b 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -3291,6 +3291,88 @@ add_unique_group_var(PlannerInfo *root, List *varinfos,
return varinfos;
}
+/*
+ * Helper routine for estimate_num_groups: add an item to a list of
+ * GroupExprInfos, but only if it's not known equal to any of the existing
+ * entries.
+ */
+typedef struct
+{
+ Node *expr; /* expression */
+ RelOptInfo *rel; /* relation it belongs to */
+ List *varinfos; /* info for variables in this expression */
+} GroupExprInfo;
+
+static List *
+add_unique_group_expr(PlannerInfo *root, List *exprinfos,
+ Node *expr, List *vars)
+{
+ GroupExprInfo *exprinfo;
+ ListCell *lc;
+ Bitmapset *varnos;
+ Index varno;
+
+ foreach(lc, exprinfos)
+ {
+ exprinfo = (GroupExprInfo *) lfirst(lc);
+
+ /* Drop exact duplicates */
+ if (equal(expr, exprinfo->expr))
+ return exprinfos;
+ }
+
+ exprinfo = (GroupExprInfo *) palloc(sizeof(GroupExprInfo));
+
+ varnos = pull_varnos(expr);
+
+ /*
+ * Expressions with vars from multiple relations should never get
+ * here, as we split them to vars.
+ */
+ Assert(bms_num_members(varnos) == 1);
+
+ varno = bms_singleton_member(varnos);
+
+ exprinfo->expr = expr;
+ exprinfo->varinfos = NIL;
+ exprinfo->rel = root->simple_rel_array[varno];
+
+ Assert(exprinfo->rel);
+
+ /* Track vars for this expression. */
+ foreach (lc, vars)
+ {
+ VariableStatData vardata;
+ Node *var = (Node *) lfirst(lc);
+
+ /* can we get no vardata for the variable? */
+ examine_variable(root, var, 0, &vardata);
+
+ exprinfo->varinfos
+ = add_unique_group_var(root, exprinfo->varinfos, var, &vardata);
+
+ ReleaseVariableStats(vardata);
+ }
+
+ /* without a list of variables, use the expression itself */
+ if (vars == NIL)
+ {
+ VariableStatData vardata;
+
+ /* can we get no vardata for the variable? */
+ examine_variable(root, expr, 0, &vardata);
+
+ exprinfo->varinfos
+ = add_unique_group_var(root, exprinfo->varinfos,
+ expr, &vardata);
+
+ ReleaseVariableStats(vardata);
+ }
+
+ return lappend(exprinfos, exprinfo);
+}
+
+
/*
* estimate_num_groups - Estimate number of groups in a grouped query
*
@@ -3360,7 +3442,7 @@ double
estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
List **pgset)
{
- List *varinfos = NIL;
+ List *exprinfos = NIL;
double srf_multiplier = 1.0;
double numdistinct;
ListCell *l;
@@ -3398,6 +3480,7 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
double this_srf_multiplier;
VariableStatData vardata;
List *varshere;
+ Relids varnos;
ListCell *l2;
/* is expression in this grouping set? */
@@ -3434,8 +3517,9 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
examine_variable(root, groupexpr, 0, &vardata);
if (HeapTupleIsValid(vardata.statsTuple) || vardata.isunique)
{
- varinfos = add_unique_group_var(root, varinfos,
- groupexpr, &vardata);
+ exprinfos = add_unique_group_expr(root, exprinfos,
+ groupexpr, NIL);
+
ReleaseVariableStats(vardata);
continue;
}
@@ -3465,6 +3549,19 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
continue;
}
+ /*
+ * Are all the variables from the same relation? If yes, search for
+ * an extended statistic matching this expression exactly.
+ */
+ varnos = pull_varnos((Node *) varshere);
+ if (bms_membership(varnos) == BMS_SINGLETON)
+ {
+ exprinfos = add_unique_group_expr(root, exprinfos,
+ groupexpr,
+ varshere);
+ continue;
+ }
+
/*
* Else add variables to varinfos list
*/
@@ -3472,9 +3569,8 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
{
Node *var = (Node *) lfirst(l2);
- examine_variable(root, var, 0, &vardata);
- varinfos = add_unique_group_var(root, varinfos, var, &vardata);
- ReleaseVariableStats(vardata);
+ exprinfos = add_unique_group_expr(root, exprinfos,
+ var, NIL);
}
}
@@ -3482,7 +3578,7 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
* If now no Vars, we must have an all-constant or all-boolean GROUP BY
* list.
*/
- if (varinfos == NIL)
+ if (exprinfos == NIL)
{
/* Apply SRF multiplier as we would do in the long path */
numdistinct *= srf_multiplier;
@@ -3506,32 +3602,32 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
*/
do
{
- GroupVarInfo *varinfo1 = (GroupVarInfo *) linitial(varinfos);
- RelOptInfo *rel = varinfo1->rel;
+ GroupExprInfo *exprinfo1 = (GroupExprInfo *) linitial(exprinfos);
+ RelOptInfo *rel = exprinfo1->rel;
double reldistinct = 1;
double relmaxndistinct = reldistinct;
int relvarcount = 0;
- List *newvarinfos = NIL;
- List *relvarinfos = NIL;
+ List *newexprinfos = NIL;
+ List *relexprinfos = NIL;
/*
* Split the list of varinfos in two - one for the current rel, one
* for remaining Vars on other rels.
*/
- relvarinfos = lappend(relvarinfos, varinfo1);
- for_each_from(l, varinfos, 1)
+ relexprinfos = lappend(relexprinfos, exprinfo1);
+ for_each_from(l, exprinfos, 1)
{
- GroupVarInfo *varinfo2 = (GroupVarInfo *) lfirst(l);
+ GroupExprInfo *exprinfo2 = (GroupExprInfo *) lfirst(l);
- if (varinfo2->rel == varinfo1->rel)
+ if (exprinfo2->rel == exprinfo1->rel)
{
/* varinfos on current rel */
- relvarinfos = lappend(relvarinfos, varinfo2);
+ relexprinfos = lappend(relexprinfos, exprinfo2);
}
else
{
- /* not time to process varinfo2 yet */
- newvarinfos = lappend(newvarinfos, varinfo2);
+ /* not time to process exprinfo2 yet */
+ newexprinfos = lappend(newexprinfos, exprinfo2);
}
}
@@ -3547,11 +3643,11 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
* apply. We apply a fudge factor below, but only if we multiplied
* more than one such values.
*/
- while (relvarinfos)
+ while (relexprinfos)
{
double mvndistinct;
- if (estimate_multivariate_ndistinct(root, rel, &relvarinfos,
+ if (estimate_multivariate_ndistinct(root, rel, &relexprinfos,
&mvndistinct))
{
reldistinct *= mvndistinct;
@@ -3561,18 +3657,24 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
}
else
{
- foreach(l, relvarinfos)
+ foreach(l, relexprinfos)
{
- GroupVarInfo *varinfo2 = (GroupVarInfo *) lfirst(l);
+ ListCell *lc;
+ GroupExprInfo *exprinfo2 = (GroupExprInfo *) lfirst(l);
- reldistinct *= varinfo2->ndistinct;
- if (relmaxndistinct < varinfo2->ndistinct)
- relmaxndistinct = varinfo2->ndistinct;
- relvarcount++;
+ foreach (lc, exprinfo2->varinfos)
+ {
+ GroupVarInfo *varinfo2 = (GroupVarInfo *) lfirst(lc);
+
+ reldistinct *= varinfo2->ndistinct;
+ if (relmaxndistinct < varinfo2->ndistinct)
+ relmaxndistinct = varinfo2->ndistinct;
+ relvarcount++;
+ }
}
/* we're done with this relation */
- relvarinfos = NIL;
+ relexprinfos = NIL;
}
}
@@ -3658,8 +3760,8 @@ estimate_num_groups(PlannerInfo *root, List *groupExprs, double input_rows,
numdistinct *= reldistinct;
}
- varinfos = newvarinfos;
- } while (varinfos != NIL);
+ exprinfos = newexprinfos;
+ } while (exprinfos != NIL);
/* Now we can account for the effects of any SRFs */
numdistinct *= srf_multiplier;
@@ -3877,53 +3979,114 @@ estimate_hashagg_tablesize(PlannerInfo *root, Path *path,
*/
static bool
estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
- List **varinfos, double *ndistinct)
+ List **exprinfos, double *ndistinct)
{
ListCell *lc;
- Bitmapset *attnums = NULL;
- int nmatches;
+ int nmatches_vars;
+ int nmatches_exprs;
Oid statOid = InvalidOid;
MVNDistinct *stats;
- Bitmapset *matched = NULL;
+ StatisticExtInfo *matched_info = NULL;
/* bail out immediately if the table has no extended statistics */
if (!rel->statlist)
return false;
- /* Determine the attnums we're looking for */
- foreach(lc, *varinfos)
- {
- GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc);
- AttrNumber attnum;
-
- Assert(varinfo->rel == rel);
-
- if (!IsA(varinfo->var, Var))
- continue;
-
- attnum = ((Var *) varinfo->var)->varattno;
-
- if (!AttrNumberIsForUserDefinedAttr(attnum))
- continue;
-
- attnums = bms_add_member(attnums, attnum);
- }
-
/* look for the ndistinct statistics matching the most vars */
- nmatches = 1; /* we require at least two matches */
+ nmatches_vars = 0; /* we require at least two matches */
+ nmatches_exprs = 0;
foreach(lc, rel->statlist)
{
+ ListCell *lc2;
StatisticExtInfo *info = (StatisticExtInfo *) lfirst(lc);
- Bitmapset *shared;
- int nshared;
+ int nshared_vars = 0;
+ int nshared_exprs = 0;
/* skip statistics of other kinds */
if (info->kind != STATS_EXT_NDISTINCT)
continue;
- /* compute attnums shared by the vars and the statistics object */
- shared = bms_intersect(info->keys, attnums);
- nshared = bms_num_members(shared);
+ /*
+ * Determine how many expressions (and variables in non-matched
+ * expressions) match. We'll then use these numbers to pick the
+ * statistics object that best matches the clauses.
+ *
+ * XXX There's a bit of trouble with expressions - we search for
+ * an exact match first, and if we don't find a match we try to
+ * search for smaller "partial" expressions extracted from it.
+ * So for example given GROUP BY (a+b) we search for statistics
+ * defined on (a+b) first, and then maybe for one on (a) and (b).
+ * The trouble here is that with the current coding, the one
+ * matching (a) and (b) might win, because we're comparing the
+ * counts. We should probably give some preference to exact
+ * matches of the expressions.
+ */
+ foreach(lc2, *exprinfos)
+ {
+ ListCell *lc3;
+ GroupExprInfo *exprinfo = (GroupExprInfo *) lfirst(lc2);
+ AttrNumber attnum;
+ bool found = false;
+
+ Assert(exprinfo->rel == rel);
+
+ /* simple Var, search in statistics keys directly */
+ if (IsA(exprinfo->expr, Var))
+ {
+ attnum = ((Var *) exprinfo->expr)->varattno;
+
+ if (!AttrNumberIsForUserDefinedAttr(attnum))
+ continue;
+
+ if (bms_is_member(attnum, info->keys))
+ nshared_vars++;
+
+ continue;
+ }
+
+ /* expression - see if it's in the statistics */
+ foreach (lc3, info->exprs)
+ {
+ Node *expr = (Node *) lfirst(lc3);
+
+ if (equal(exprinfo->expr, expr))
+ {
+ nshared_exprs++;
+ found = true;
+ break;
+ }
+ }
+
+ /*
+ * If it's a complex expression, and we have found it in the
+ * statistics object, we're done. Otherwise try to match the
+ * varinfos we've extracted from the expression. That way we
+ * can do at least some estimation.
+ */
+ if (found)
+ continue;
+
+ foreach(lc3, exprinfo->varinfos)
+ {
+ GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc3);
+
+ if (IsA(varinfo->var, Var))
+ {
+ attnum = ((Var *) exprinfo->expr)->varattno;
+
+ if (!AttrNumberIsForUserDefinedAttr(attnum))
+ continue;
+
+ if (bms_is_member(attnum, info->keys))
+ nshared_vars++;
+ }
+
+ /* XXX What if it's not a Var? */
+ }
+ }
+
+ if (nshared_vars + nshared_exprs < 2)
+ continue;
/*
* Does this statistics object match more columns than the currently
@@ -3932,18 +4095,21 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
* XXX This should break ties using name of the object, or something
* like that, to make the outcome stable.
*/
- if (nshared > nmatches)
+ if ((nshared_vars > nmatches_vars) ||
+ ((nshared_vars == nmatches_vars) && (nshared_exprs > nmatches_exprs)))
{
statOid = info->statOid;
- nmatches = nshared;
- matched = shared;
+ nmatches_vars = nshared_vars;
+ nmatches_exprs = nshared_exprs;
+ matched_info = info;
}
}
/* No match? */
if (statOid == InvalidOid)
return false;
- Assert(nmatches > 1 && matched != NULL);
+
+ Assert(nmatches_vars + nmatches_exprs > 1);
stats = statext_ndistinct_load(statOid);
@@ -3956,6 +4122,56 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
int i;
List *newlist = NIL;
MVNDistinctItem *item = NULL;
+ ListCell *lc2;
+ Bitmapset *matched = NULL;
+
+ /* see what actually matched */
+ foreach (lc2, *exprinfos)
+ {
+ ListCell *lc3;
+ int idx;
+ bool found = false;
+
+ GroupExprInfo *exprinfo = (GroupExprInfo *) lfirst(lc2);
+
+ /* expression - see if it's in the statistics */
+ idx = 0;
+ foreach (lc3, matched_info->exprs)
+ {
+ Node *expr = (Node *) lfirst(lc3);
+
+ idx++;
+
+ if (equal(exprinfo->expr, expr))
+ {
+ matched = bms_add_member(matched, MaxHeapAttributeNumber + idx);
+ found = true;
+ break;
+ }
+ }
+
+ if (found)
+ continue;
+
+ foreach (lc3, exprinfo->varinfos)
+ {
+ GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc2);
+
+ /* simple Var, search in statistics keys directly */
+ if (IsA(varinfo->var, Var))
+ {
+ AttrNumber attnum = ((Var *) varinfo->var)->varattno;
+
+ if (!AttrNumberIsForUserDefinedAttr(attnum))
+ continue;
+
+ if (!bms_is_member(attnum, matched_info->keys))
+ continue;
+
+ matched = bms_add_member(matched, attnum);
+ }
+ }
+ }
/* Find the specific item that exactly matches the combination */
for (i = 0; i < stats->nitems; i++)
@@ -3973,28 +4189,49 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
if (!item)
elog(ERROR, "corrupt MVNDistinct entry");
- /* Form the output varinfo list, keeping only unmatched ones */
- foreach(lc, *varinfos)
+ /* Form the output exprinfo list, keeping only unmatched ones */
+ foreach(lc, *exprinfos)
{
- GroupVarInfo *varinfo = (GroupVarInfo *) lfirst(lc);
+ GroupExprInfo *exprinfo = (GroupExprInfo *) lfirst(lc);
AttrNumber attnum;
+ ListCell *lc3;
+ bool found = false;
+
+ foreach (lc3, matched_info->exprs)
+ {
+ Node *expr = (Node *) lfirst(lc3);
+
+ if (equal(exprinfo->expr, expr))
+ {
+ found = true;
+ break;
+ }
+ }
- if (!IsA(varinfo->var, Var))
+ /* the whole expression was matched, so skip it */
+ if (found)
+ continue;
+
+ if (!IsA(exprinfo->expr, Var))
{
- newlist = lappend(newlist, varinfo);
+ /*
+ * FIXME Probably should remove varinfos that match the
+ * selected MVNDistinct item.
+ */
+ newlist = lappend(newlist, exprinfo);
continue;
}
- attnum = ((Var *) varinfo->var)->varattno;
+ attnum = ((Var *) exprinfo->expr)->varattno;
if (!AttrNumberIsForUserDefinedAttr(attnum))
continue;
if (!bms_is_member(attnum, matched))
- newlist = lappend(newlist, varinfo);
+ newlist = lappend(newlist, exprinfo);
}
- *varinfos = newlist;
+ *exprinfos = newlist;
*ndistinct = item->ndistinct;
return true;
}
@@ -4690,6 +4927,13 @@ get_join_variables(PlannerInfo *root, List *args, SpecialJoinInfo *sjinfo,
*join_is_reversed = false;
}
+/* statext_expressions_load copies the tuple, so just pfree it. */
+static void
+ReleaseDummy(HeapTuple tuple)
+{
+ pfree(tuple);
+}
+
/*
* examine_variable
* Try to look up statistical data about an expression.
@@ -4830,6 +5074,7 @@ examine_variable(PlannerInfo *root, Node *node, int varRelid,
* operator we are estimating for. FIXME later.
*/
ListCell *ilist;
+ ListCell *slist;
foreach(ilist, onerel->indexlist)
{
@@ -4986,6 +5231,68 @@ examine_variable(PlannerInfo *root, Node *node, int varRelid,
if (vardata->statsTuple)
break;
}
+
+ /*
+ * Search extended statistics for one with a matching expression.
+ * There might be multiple ones, so just grab the first one. In
+ * the future, we might consider the statistics target (and pick
+ * the most accurate statistics) and maybe some other parameters.
+ */
+ foreach(slist, onerel->statlist)
+ {
+ StatisticExtInfo *info = (StatisticExtInfo *) lfirst(slist);
+ ListCell *expr_item;
+ int pos;
+
+ /*
+ * Stop once we've found statistics for the expression (either
+ * from extended stats, or for an index in the preceding loop).
+ */
+ if (vardata->statsTuple)
+ break;
+
+ /* skip stats without per-expression stats */
+ if (info->kind != STATS_EXT_EXPRESSIONS)
+ continue;
+
+ pos = 0;
+ foreach (expr_item, info->exprs)
+ {
+ Node *expr = (Node *) lfirst(expr_item);
+
+ Assert(expr);
+
+ /* strip RelabelType before comparing it */
+ if (expr && IsA(expr, RelabelType))
+ expr = (Node *) ((RelabelType *) expr)->arg;
+
+ /* found a match, see if we can extract pg_statistic row */
+ if (equal(node, expr))
+ {
+ HeapTuple t = statext_expressions_load(info->statOid, pos);
+
+ vardata->statsTuple = t;
+
+ /*
+ * FIXME not sure if we should cache the tuple somewhere?
+ * It's stored in a cached tuple in the "data" catalog,
+ * and we just create a new copy every time.
+ */
+ vardata->freefunc = ReleaseDummy;
+
+ /*
+ * FIXME Hack to make statistic_proc_security_check happy,
+ * so that this does not get rejected. Probably needs more
+ * thought, just a hack.
+ */
+ vardata->acl_ok = true;
+
+ break;
+ }
+
+ pos++;
+ }
+ }
}
}
diff --git a/src/bin/pg_dump/t/002_pg_dump.pl b/src/bin/pg_dump/t/002_pg_dump.pl
index 2b501166b8..23cc23d037 100644
--- a/src/bin/pg_dump/t/002_pg_dump.pl
+++ b/src/bin/pg_dump/t/002_pg_dump.pl
@@ -2591,6 +2591,18 @@ my %tests = (
unlike => { exclude_dump_test_schema => 1, },
},
+ 'CREATE STATISTICS extended_stats_expression' => {
+ create_order => 99,
+ create_sql => 'CREATE STATISTICS dump_test.test_ext_stats_expr
+ ON (2 * col1) FROM dump_test.test_fifth_table',
+ regexp => qr/^
+ \QCREATE STATISTICS dump_test.test_ext_stats_expr ON ((2 * col1)) FROM dump_test.test_fifth_table;\E
+ /xms,
+ like =>
+ { %full_runs, %dump_test_schema_runs, section_post_data => 1, },
+ unlike => { exclude_dump_test_schema => 1, },
+ },
+
'CREATE SEQUENCE test_table_col1_seq' => {
regexp => qr/^
\QCREATE SEQUENCE dump_test.test_table_col1_seq\E
diff --git a/src/bin/psql/describe.c b/src/bin/psql/describe.c
index 20af5a92b4..c1333b19d6 100644
--- a/src/bin/psql/describe.c
+++ b/src/bin/psql/describe.c
@@ -2680,15 +2680,16 @@ describeOneTableDetails(const char *schemaname,
/* print any extended statistics */
if (pset.sversion >= 100000)
{
+ /*
+ * FIXME this needs to be version-dependent, because older
+ * versions don't have pg_get_statisticsobjdef_columns.
+ */
printfPQExpBuffer(&buf,
"SELECT oid, "
"stxrelid::pg_catalog.regclass, "
"stxnamespace::pg_catalog.regnamespace AS nsp, "
"stxname,\n"
- " (SELECT pg_catalog.string_agg(pg_catalog.quote_ident(attname),', ')\n"
- " FROM pg_catalog.unnest(stxkeys) s(attnum)\n"
- " JOIN pg_catalog.pg_attribute a ON (stxrelid = a.attrelid AND\n"
- " a.attnum = s.attnum AND NOT attisdropped)) AS columns,\n"
+ "pg_get_statisticsobjdef_columns(oid) AS columns,\n"
" 'd' = any(stxkind) AS ndist_enabled,\n"
" 'f' = any(stxkind) AS deps_enabled,\n"
" 'm' = any(stxkind) AS mcv_enabled,\n");
@@ -2715,33 +2716,60 @@ describeOneTableDetails(const char *schemaname,
for (i = 0; i < tuples; i++)
{
bool gotone = false;
+ bool has_ndistinct;
+ bool has_dependencies;
+ bool has_mcv;
+ bool has_all;
+ bool has_some;
+
+ has_ndistinct = (strcmp(PQgetvalue(result, i, 5), "t") == 0);
+ has_dependencies = (strcmp(PQgetvalue(result, i, 6), "t") == 0);
+ has_mcv = (strcmp(PQgetvalue(result, i, 7), "t") == 0);
printfPQExpBuffer(&buf, " ");
/* statistics object name (qualified with namespace) */
- appendPQExpBuffer(&buf, "\"%s\".\"%s\" (",
+ appendPQExpBuffer(&buf, "\"%s\".\"%s\"",
PQgetvalue(result, i, 2),
PQgetvalue(result, i, 3));
- /* options */
- if (strcmp(PQgetvalue(result, i, 5), "t") == 0)
- {
- appendPQExpBufferStr(&buf, "ndistinct");
- gotone = true;
- }
+ /*
+ * When printing kinds we ignore expression statistics, which
+ * is used only internally and can't be specified by user.
+ * We don't print the kinds when either none are specified
+ * (in which case it has to be statistics on a single expr)
+ * or when all are specified (in which case we assume it's
+ * expanded by CREATE STATISTICS).
+ */
+ has_all = (has_ndistinct && has_dependencies && has_mcv);
+ has_some = (has_ndistinct || has_dependencies || has_mcv);
- if (strcmp(PQgetvalue(result, i, 6), "t") == 0)
+ if (has_some && !has_all)
{
- appendPQExpBuffer(&buf, "%sdependencies", gotone ? ", " : "");
- gotone = true;
- }
+ appendPQExpBuffer(&buf, " (");
- if (strcmp(PQgetvalue(result, i, 7), "t") == 0)
- {
- appendPQExpBuffer(&buf, "%smcv", gotone ? ", " : "");
+ /* options */
+ if (has_ndistinct)
+ {
+ appendPQExpBufferStr(&buf, "ndistinct");
+ gotone = true;
+ }
+
+ if (has_dependencies)
+ {
+ appendPQExpBuffer(&buf, "%sdependencies", gotone ? ", " : "");
+ gotone = true;
+ }
+
+ if (has_mcv)
+ {
+ appendPQExpBuffer(&buf, "%smcv", gotone ? ", " : "");
+ }
+
+ appendPQExpBuffer(&buf, ")");
}
- appendPQExpBuffer(&buf, ") ON %s FROM %s",
+ appendPQExpBuffer(&buf, " ON %s FROM %s",
PQgetvalue(result, i, 4),
PQgetvalue(result, i, 1));
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index b5f52d4e4a..ff33e2f960 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -3652,6 +3652,14 @@
proname => 'pg_get_statisticsobjdef', provolatile => 's',
prorettype => 'text', proargtypes => 'oid',
prosrc => 'pg_get_statisticsobjdef' },
+{ oid => '8887', descr => 'extended statistics columns',
+ proname => 'pg_get_statisticsobjdef_columns', provolatile => 's',
+ prorettype => 'text', proargtypes => 'oid',
+ prosrc => 'pg_get_statisticsobjdef_columns' },
+{ oid => '8886', descr => 'extended statistics expressions',
+ proname => 'pg_get_statisticsobjdef_expressions', provolatile => 's',
+ prorettype => '_text', proargtypes => 'oid',
+ prosrc => 'pg_get_statisticsobjdef_expressions' },
{ oid => '3352', descr => 'partition key description',
proname => 'pg_get_partkeydef', provolatile => 's', prorettype => 'text',
proargtypes => 'oid', prosrc => 'pg_get_partkeydef' },
diff --git a/src/include/catalog/pg_statistic_ext.h b/src/include/catalog/pg_statistic_ext.h
index 99f6cea0a5..cf46a79af9 100644
--- a/src/include/catalog/pg_statistic_ext.h
+++ b/src/include/catalog/pg_statistic_ext.h
@@ -52,6 +52,9 @@ CATALOG(pg_statistic_ext,3381,StatisticExtRelationId)
#ifdef CATALOG_VARLEN
char stxkind[1] BKI_FORCE_NOT_NULL; /* statistics kinds requested
* to build */
+ pg_node_tree stxexprs; /* A list of expression trees for stats
+ * attributes that are not simple column
+ * references. */
#endif
} FormData_pg_statistic_ext;
@@ -77,6 +80,7 @@ DECLARE_INDEX(pg_statistic_ext_relid_index, 3379, on pg_statistic_ext using btre
#define STATS_EXT_NDISTINCT 'd'
#define STATS_EXT_DEPENDENCIES 'f'
#define STATS_EXT_MCV 'm'
+#define STATS_EXT_EXPRESSIONS 'e'
#endif /* EXPOSE_TO_CLIENT_CODE */
diff --git a/src/include/catalog/pg_statistic_ext_data.h b/src/include/catalog/pg_statistic_ext_data.h
index e0aa152f7b..0d2f6a6c32 100644
--- a/src/include/catalog/pg_statistic_ext_data.h
+++ b/src/include/catalog/pg_statistic_ext_data.h
@@ -37,6 +37,7 @@ CATALOG(pg_statistic_ext_data,3429,StatisticExtDataRelationId)
pg_ndistinct stxdndistinct; /* ndistinct coefficients (serialized) */
pg_dependencies stxddependencies; /* dependencies (serialized) */
pg_mcv_list stxdmcv; /* MCV (serialized) */
+ pg_statistic stxdexpr[1]; /* stats for expressions */
#endif
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index caed683ba9..374f047dda 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -451,6 +451,7 @@ typedef enum NodeTag
T_TypeName,
T_ColumnDef,
T_IndexElem,
+ T_StatsElem,
T_Constraint,
T_DefElem,
T_RangeTblEntry,
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index dc2bb40926..f2042ba445 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -2830,8 +2830,24 @@ typedef struct CreateStatsStmt
List *relations; /* rels to build stats on (list of RangeVar) */
char *stxcomment; /* comment to apply to stats, or NULL */
bool if_not_exists; /* do nothing if stats name already exists */
+ bool transformed; /* true when transformStatsStmt is finished */
} CreateStatsStmt;
+/*
+ * StatsElem - statistics parameters (used in CREATE STATISTICS)
+ *
+ * For a plain attribute, 'name' is the name of the referenced table column
+ * and 'expr' is NULL. For an expression, 'name' is NULL and 'expr' is the
+ * expression tree.
+ */
+typedef struct StatsElem
+{
+ NodeTag type;
+ char *name; /* name of attribute to index, or NULL */
+ Node *expr; /* expression to index, or NULL */
+} StatsElem;
+
+
/* ----------------------
* Alter Statistics Statement
* ----------------------
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index cde2637798..c384f2c6e7 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -915,8 +915,9 @@ typedef struct StatisticExtInfo
Oid statOid; /* OID of the statistics row */
RelOptInfo *rel; /* back-link to statistic's table */
- char kind; /* statistic kind of this entry */
+ char kind; /* statistics kind of this entry */
Bitmapset *keys; /* attnums of the columns covered */
+ List *exprs; /* expressions */
} StatisticExtInfo;
/*
diff --git a/src/include/parser/parse_node.h b/src/include/parser/parse_node.h
index dfc214b06f..2b477c38eb 100644
--- a/src/include/parser/parse_node.h
+++ b/src/include/parser/parse_node.h
@@ -69,6 +69,7 @@ typedef enum ParseExprKind
EXPR_KIND_FUNCTION_DEFAULT, /* default parameter value for function */
EXPR_KIND_INDEX_EXPRESSION, /* index expression */
EXPR_KIND_INDEX_PREDICATE, /* index predicate */
+ EXPR_KIND_STATS_EXPRESSION, /* extended statistics expression */
EXPR_KIND_ALTER_COL_TRANSFORM, /* transform expr in ALTER COLUMN TYPE */
EXPR_KIND_EXECUTE_PARAMETER, /* parameter value in EXECUTE */
EXPR_KIND_TRIGGER_WHEN, /* WHEN condition in CREATE TRIGGER */
diff --git a/src/include/parser/parse_utilcmd.h b/src/include/parser/parse_utilcmd.h
index bfa4a6b0f2..1056bf081b 100644
--- a/src/include/parser/parse_utilcmd.h
+++ b/src/include/parser/parse_utilcmd.h
@@ -26,6 +26,8 @@ extern AlterTableStmt *transformAlterTableStmt(Oid relid, AlterTableStmt *stmt,
List **afterStmts);
extern IndexStmt *transformIndexStmt(Oid relid, IndexStmt *stmt,
const char *queryString);
+extern CreateStatsStmt *transformStatsStmt(Oid relid, CreateStatsStmt *stmt,
+ const char *queryString);
extern void transformRuleStmt(RuleStmt *stmt, const char *queryString,
List **actions, Node **whereClause);
extern List *transformCreateSchemaStmt(CreateSchemaStmt *stmt);
diff --git a/src/include/statistics/extended_stats_internal.h b/src/include/statistics/extended_stats_internal.h
index c849bd57c0..092bc3eb8a 100644
--- a/src/include/statistics/extended_stats_internal.h
+++ b/src/include/statistics/extended_stats_internal.h
@@ -57,19 +57,35 @@ typedef struct SortItem
int count;
} SortItem;
+/*
+ * Used to pass pre-computed information about expressions the stats
+ * object is defined on.
+ */
+typedef struct ExprInfo
+{
+ int nexprs; /* number of expressions */
+ Oid *collations; /* collation for each expression */
+ Oid *types; /* type of each expression */
+ Datum **values; /* values for each expression */
+ bool **nulls; /* nulls for each expression */
+} ExprInfo;
+
extern MVNDistinct *statext_ndistinct_build(double totalrows,
int numrows, HeapTuple *rows,
- Bitmapset *attrs, VacAttrStats **stats);
+ ExprInfo *exprs, Bitmapset *attrs,
+ VacAttrStats **stats);
extern bytea *statext_ndistinct_serialize(MVNDistinct *ndistinct);
extern MVNDistinct *statext_ndistinct_deserialize(bytea *data);
extern MVDependencies *statext_dependencies_build(int numrows, HeapTuple *rows,
- Bitmapset *attrs, VacAttrStats **stats);
+ ExprInfo *exprs, Bitmapset *attrs,
+ VacAttrStats **stats);
extern bytea *statext_dependencies_serialize(MVDependencies *dependencies);
extern MVDependencies *statext_dependencies_deserialize(bytea *data);
extern MCVList *statext_mcv_build(int numrows, HeapTuple *rows,
- Bitmapset *attrs, VacAttrStats **stats,
+ ExprInfo *exprs, Bitmapset *attrs,
+ VacAttrStats **stats,
double totalrows, int stattarget);
extern bytea *statext_mcv_serialize(MCVList *mcv, VacAttrStats **stats);
extern MCVList *statext_mcv_deserialize(bytea *data);
@@ -93,11 +109,18 @@ extern void *bsearch_arg(const void *key, const void *base,
extern AttrNumber *build_attnums_array(Bitmapset *attrs, int *numattrs);
extern SortItem *build_sorted_items(int numrows, int *nitems, HeapTuple *rows,
- TupleDesc tdesc, MultiSortSupport mss,
+ ExprInfo *exprs, TupleDesc tdesc,
+ MultiSortSupport mss,
int numattrs, AttrNumber *attnums);
extern bool examine_clause_args(List *args, Var **varp,
Const **cstp, bool *varonleftp);
+extern bool examine_clause_args2(List *args, Node **exprp,
+ Const **cstp, bool *expronleftp);
+extern bool examine_opclause_expression(OpExpr *expr, Var **varp, Const **cstp,
+ bool *varonleftp);
+extern bool examine_opclause_expression2(OpExpr *expr, Node **exprp, Const **cstp,
+ bool *expronleftp);
extern Selectivity mcv_combine_selectivities(Selectivity simple_sel,
Selectivity mcv_sel,
@@ -124,4 +147,13 @@ extern Selectivity mcv_clause_selectivity_or(PlannerInfo *root,
Selectivity *overlap_basesel,
Selectivity *totalsel);
+extern Bitmapset *add_expressions_to_attributes(Bitmapset *attrs, int nexprs);
+
+/* translate 0-based expression index to attnum and back */
+#define EXPRESSION_ATTNUM(index) \
+ (MaxHeapAttributeNumber + (index) + 1)
+
+#define EXPRESSION_INDEX(attnum) \
+ ((attnum) - MaxHeapAttributeNumber - 1)
+
#endif /* EXTENDED_STATS_INTERNAL_H */
diff --git a/src/include/statistics/statistics.h b/src/include/statistics/statistics.h
index fec50688ea..006d578e0c 100644
--- a/src/include/statistics/statistics.h
+++ b/src/include/statistics/statistics.h
@@ -121,6 +121,8 @@ extern Selectivity statext_clauselist_selectivity(PlannerInfo *root,
extern bool has_stats_of_kind(List *stats, char requiredkind);
extern StatisticExtInfo *choose_best_statistics(List *stats, char requiredkind,
Bitmapset **clause_attnums,
+ List **clause_exprs,
int nclauses);
+extern HeapTuple statext_expressions_load(Oid stxoid, int idx);
#endif /* STATISTICS_H */
diff --git a/src/test/regress/expected/create_table_like.out b/src/test/regress/expected/create_table_like.out
index 10d17be23c..4dc5e6aa5f 100644
--- a/src/test/regress/expected/create_table_like.out
+++ b/src/test/regress/expected/create_table_like.out
@@ -304,7 +304,9 @@ CREATE TABLE ctlt1 (a text CHECK (length(a) > 2) PRIMARY KEY, b text);
CREATE INDEX ctlt1_b_key ON ctlt1 (b);
CREATE INDEX ctlt1_fnidx ON ctlt1 ((a || b));
CREATE STATISTICS ctlt1_a_b_stat ON a,b FROM ctlt1;
+CREATE STATISTICS ctlt1_expr_stat ON (a || b) FROM ctlt1;
COMMENT ON STATISTICS ctlt1_a_b_stat IS 'ab stats';
+COMMENT ON STATISTICS ctlt1_expr_stat IS 'ab expr stats';
COMMENT ON COLUMN ctlt1.a IS 'A';
COMMENT ON COLUMN ctlt1.b IS 'B';
COMMENT ON CONSTRAINT ctlt1_a_check ON ctlt1 IS 't1_a_check';
@@ -414,7 +416,8 @@ Indexes:
Check constraints:
"ctlt1_a_check" CHECK (length(a) > 2)
Statistics objects:
- "public"."ctlt_all_a_b_stat" (ndistinct, dependencies, mcv) ON a, b FROM ctlt_all
+ "public"."ctlt_all_a_b_stat" ON a, b FROM ctlt_all
+ "public"."ctlt_all_expr_stat" ON ((a || b)) FROM ctlt_all
SELECT c.relname, objsubid, description FROM pg_description, pg_index i, pg_class c WHERE classoid = 'pg_class'::regclass AND objoid = i.indexrelid AND c.oid = i.indexrelid AND i.indrelid = 'ctlt_all'::regclass ORDER BY c.relname, objsubid;
relname | objsubid | description
@@ -424,10 +427,11 @@ SELECT c.relname, objsubid, description FROM pg_description, pg_index i, pg_clas
(2 rows)
SELECT s.stxname, objsubid, description FROM pg_description, pg_statistic_ext s WHERE classoid = 'pg_statistic_ext'::regclass AND objoid = s.oid AND s.stxrelid = 'ctlt_all'::regclass ORDER BY s.stxname, objsubid;
- stxname | objsubid | description
--------------------+----------+-------------
- ctlt_all_a_b_stat | 0 | ab stats
-(1 row)
+ stxname | objsubid | description
+--------------------+----------+---------------
+ ctlt_all_a_b_stat | 0 | ab stats
+ ctlt_all_expr_stat | 0 | ab expr stats
+(2 rows)
CREATE TABLE inh_error1 () INHERITS (ctlt1, ctlt4);
NOTICE: merging multiple inherited definitions of column "a"
@@ -452,7 +456,8 @@ Indexes:
Check constraints:
"ctlt1_a_check" CHECK (length(a) > 2)
Statistics objects:
- "public"."pg_attrdef_a_b_stat" (ndistinct, dependencies, mcv) ON a, b FROM public.pg_attrdef
+ "public"."pg_attrdef_a_b_stat" ON a, b FROM public.pg_attrdef
+ "public"."pg_attrdef_expr_stat" ON ((a || b)) FROM public.pg_attrdef
DROP TABLE public.pg_attrdef;
-- Check that LIKE isn't confused when new table masks the old, either
@@ -473,7 +478,8 @@ Indexes:
Check constraints:
"ctlt1_a_check" CHECK (length(a) > 2)
Statistics objects:
- "ctl_schema"."ctlt1_a_b_stat" (ndistinct, dependencies, mcv) ON a, b FROM ctlt1
+ "ctl_schema"."ctlt1_a_b_stat" ON a, b FROM ctlt1
+ "ctl_schema"."ctlt1_expr_stat" ON ((a || b)) FROM ctlt1
ROLLBACK;
DROP TABLE ctlt1, ctlt2, ctlt3, ctlt4, ctlt12_storage, ctlt12_comments, ctlt1_inh, ctlt13_inh, ctlt13_like, ctlt_all, ctla, ctlb CASCADE;
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 6173473de9..246e151fa0 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2400,6 +2400,7 @@ pg_stats_ext| SELECT cn.nspname AS schemaname,
( SELECT array_agg(a.attname ORDER BY a.attnum) AS array_agg
FROM (unnest(s.stxkeys) k(k)
JOIN pg_attribute a ON (((a.attrelid = s.stxrelid) AND (a.attnum = k.k))))) AS attnames,
+ pg_get_statisticsobjdef_expressions(s.oid) AS exprs,
s.stxkind AS kinds,
sd.stxdndistinct AS n_distinct,
sd.stxddependencies AS dependencies,
@@ -2421,6 +2422,80 @@ pg_stats_ext| SELECT cn.nspname AS schemaname,
FROM (unnest(s.stxkeys) k(k)
JOIN pg_attribute a ON (((a.attrelid = s.stxrelid) AND (a.attnum = k.k))))
WHERE (NOT has_column_privilege(c.oid, a.attnum, 'select'::text))))) AND ((c.relrowsecurity = false) OR (NOT row_security_active(c.oid))));
+pg_stats_ext_exprs| SELECT cn.nspname AS schemaname,
+ c.relname AS tablename,
+ sn.nspname AS statistics_schemaname,
+ s.stxname AS statistics_name,
+ pg_get_userbyid(s.stxowner) AS statistics_owner,
+ stat.expr,
+ (stat.a).stanullfrac AS null_frac,
+ (stat.a).stawidth AS avg_width,
+ (stat.a).stadistinct AS n_distinct,
+ CASE
+ WHEN ((stat.a).stakind1 = 1) THEN (stat.a).stavalues1
+ WHEN ((stat.a).stakind2 = 1) THEN (stat.a).stavalues2
+ WHEN ((stat.a).stakind3 = 1) THEN (stat.a).stavalues3
+ WHEN ((stat.a).stakind4 = 1) THEN (stat.a).stavalues4
+ WHEN ((stat.a).stakind5 = 1) THEN (stat.a).stavalues5
+ ELSE NULL::anyarray
+ END AS most_common_vals,
+ CASE
+ WHEN ((stat.a).stakind1 = 1) THEN (stat.a).stanumbers1
+ WHEN ((stat.a).stakind2 = 1) THEN (stat.a).stanumbers2
+ WHEN ((stat.a).stakind3 = 1) THEN (stat.a).stanumbers3
+ WHEN ((stat.a).stakind4 = 1) THEN (stat.a).stanumbers4
+ WHEN ((stat.a).stakind5 = 1) THEN (stat.a).stanumbers5
+ ELSE NULL::real[]
+ END AS most_common_freqs,
+ CASE
+ WHEN ((stat.a).stakind1 = 2) THEN (stat.a).stavalues1
+ WHEN ((stat.a).stakind2 = 2) THEN (stat.a).stavalues2
+ WHEN ((stat.a).stakind3 = 2) THEN (stat.a).stavalues3
+ WHEN ((stat.a).stakind4 = 2) THEN (stat.a).stavalues4
+ WHEN ((stat.a).stakind5 = 2) THEN (stat.a).stavalues5
+ ELSE NULL::anyarray
+ END AS histogram_bounds,
+ CASE
+ WHEN ((stat.a).stakind1 = 3) THEN (stat.a).stanumbers1[1]
+ WHEN ((stat.a).stakind2 = 3) THEN (stat.a).stanumbers2[1]
+ WHEN ((stat.a).stakind3 = 3) THEN (stat.a).stanumbers3[1]
+ WHEN ((stat.a).stakind4 = 3) THEN (stat.a).stanumbers4[1]
+ WHEN ((stat.a).stakind5 = 3) THEN (stat.a).stanumbers5[1]
+ ELSE NULL::real
+ END AS correlation,
+ CASE
+ WHEN ((stat.a).stakind1 = 4) THEN (stat.a).stavalues1
+ WHEN ((stat.a).stakind2 = 4) THEN (stat.a).stavalues2
+ WHEN ((stat.a).stakind3 = 4) THEN (stat.a).stavalues3
+ WHEN ((stat.a).stakind4 = 4) THEN (stat.a).stavalues4
+ WHEN ((stat.a).stakind5 = 4) THEN (stat.a).stavalues5
+ ELSE NULL::anyarray
+ END AS most_common_elems,
+ CASE
+ WHEN ((stat.a).stakind1 = 4) THEN (stat.a).stanumbers1
+ WHEN ((stat.a).stakind2 = 4) THEN (stat.a).stanumbers2
+ WHEN ((stat.a).stakind3 = 4) THEN (stat.a).stanumbers3
+ WHEN ((stat.a).stakind4 = 4) THEN (stat.a).stanumbers4
+ WHEN ((stat.a).stakind5 = 4) THEN (stat.a).stanumbers5
+ ELSE NULL::real[]
+ END AS most_common_elem_freqs,
+ CASE
+ WHEN ((stat.a).stakind1 = 5) THEN (stat.a).stanumbers1
+ WHEN ((stat.a).stakind2 = 5) THEN (stat.a).stanumbers2
+ WHEN ((stat.a).stakind3 = 5) THEN (stat.a).stanumbers3
+ WHEN ((stat.a).stakind4 = 5) THEN (stat.a).stanumbers4
+ WHEN ((stat.a).stakind5 = 5) THEN (stat.a).stanumbers5
+ ELSE NULL::real[]
+ END AS elem_count_histogram
+ FROM (((((pg_statistic_ext s
+ JOIN pg_class c ON ((c.oid = s.stxrelid)))
+ JOIN pg_statistic_ext_data sd ON ((s.oid = sd.stxoid)))
+ LEFT JOIN pg_namespace cn ON ((cn.oid = c.relnamespace)))
+ LEFT JOIN pg_namespace sn ON ((sn.oid = s.stxnamespace)))
+ LEFT JOIN LATERAL ( SELECT x.expr,
+ x.a
+ FROM ( SELECT unnest(pg_get_statisticsobjdef_expressions(s.oid)) AS expr,
+ unnest(sd.stxdexpr) AS a) x) stat ON ((sd.stxdexpr IS NOT NULL)));
pg_tables| SELECT n.nspname AS schemaname,
c.relname AS tablename,
pg_get_userbyid(c.relowner) AS tableowner,
diff --git a/src/test/regress/expected/stats_ext.out b/src/test/regress/expected/stats_ext.out
index 431b3fa3de..36b7e3e7d3 100644
--- a/src/test/regress/expected/stats_ext.out
+++ b/src/test/regress/expected/stats_ext.out
@@ -25,7 +25,7 @@ begin
end;
$$;
-- Verify failures
-CREATE TABLE ext_stats_test (x int, y int, z int);
+CREATE TABLE ext_stats_test (x text, y int, z int);
CREATE STATISTICS tst;
ERROR: syntax error at or near ";"
LINE 1: CREATE STATISTICS tst;
@@ -44,12 +44,25 @@ CREATE STATISTICS tst ON a, b FROM ext_stats_test;
ERROR: column "a" does not exist
CREATE STATISTICS tst ON x, x, y FROM ext_stats_test;
ERROR: duplicate column name in statistics definition
-CREATE STATISTICS tst ON x + y FROM ext_stats_test;
-ERROR: only simple column references are allowed in CREATE STATISTICS
-CREATE STATISTICS tst ON (x, y) FROM ext_stats_test;
-ERROR: only simple column references are allowed in CREATE STATISTICS
+CREATE STATISTICS tst ON x, x, y, x, x, y, x, x, y FROM ext_stats_test;
+ERROR: cannot have more than 8 columns in statistics
+CREATE STATISTICS tst ON x, x, y, x, x, (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1) FROM ext_stats_test;
+ERROR: cannot have more than 8 columns in statistics
+CREATE STATISTICS tst ON (x || 'x'), (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1) FROM ext_stats_test;
+ERROR: cannot have more than 8 columns in statistics
+CREATE STATISTICS tst ON (x || 'x'), (x || 'x'), y FROM ext_stats_test;
+ERROR: duplicate expression in statistics definition
CREATE STATISTICS tst (unrecognized) ON x, y FROM ext_stats_test;
ERROR: unrecognized statistics kind "unrecognized"
+-- incorrect expressions
+CREATE STATISTICS tst ON y + z FROM ext_stats_test; -- missing parentheses
+ERROR: syntax error at or near "+"
+LINE 1: CREATE STATISTICS tst ON y + z FROM ext_stats_test;
+ ^
+CREATE STATISTICS tst ON (x, y) FROM ext_stats_test; -- tuple expression
+ERROR: syntax error at or near ","
+LINE 1: CREATE STATISTICS tst ON (x, y) FROM ext_stats_test;
+ ^
DROP TABLE ext_stats_test;
-- Ensure stats are dropped sanely, and test IF NOT EXISTS while at it
CREATE TABLE ab1 (a INTEGER, b INTEGER, c INTEGER);
@@ -79,7 +92,7 @@ ALTER TABLE ab1 DROP COLUMN a;
b | integer | | |
c | integer | | |
Statistics objects:
- "public"."ab1_b_c_stats" (ndistinct, dependencies, mcv) ON b, c FROM ab1
+ "public"."ab1_b_c_stats" ON b, c FROM ab1
-- Ensure statistics are dropped when table is
SELECT stxname FROM pg_statistic_ext WHERE stxname LIKE 'ab1%';
@@ -111,7 +124,7 @@ ALTER STATISTICS ab1_a_b_stats SET STATISTICS 0;
a | integer | | |
b | integer | | |
Statistics objects:
- "public"."ab1_a_b_stats" (ndistinct, dependencies, mcv) ON a, b FROM ab1; STATISTICS 0
+ "public"."ab1_a_b_stats" ON a, b FROM ab1; STATISTICS 0
ANALYZE ab1;
SELECT stxname, stxdndistinct, stxddependencies, stxdmcv
@@ -131,7 +144,7 @@ ALTER STATISTICS ab1_a_b_stats SET STATISTICS -1;
a | integer | | | | plain | |
b | integer | | | | plain | |
Statistics objects:
- "public"."ab1_a_b_stats" (ndistinct, dependencies, mcv) ON a, b FROM ab1
+ "public"."ab1_a_b_stats" ON a, b FROM ab1
-- partial analyze doesn't build stats either
ANALYZE ab1 (a);
@@ -150,6 +163,39 @@ CREATE STATISTICS ab1_a_b_stats ON a, b FROM ab1;
ANALYZE ab1;
DROP TABLE ab1 CASCADE;
NOTICE: drop cascades to table ab1c
+-- basic test for statistics on expressions
+CREATE TABLE ab1 (a INTEGER, b INTEGER, c TIMESTAMP, d TIMESTAMPTZ);
+-- expression stats may be built on a single expression column
+CREATE STATISTICS ab1_exprstat_1 ON (a+b) FROM ab1;
+-- with a single expression, we only enable expression statistics
+CREATE STATISTICS ab1_exprstat_2 ON (a+b) FROM ab1;
+SELECT stxkind FROM pg_statistic_ext WHERE stxname = 'ab1_exprstat_2';
+ stxkind
+---------
+ {e}
+(1 row)
+
+-- adding anything to the expression builds all statistics kinds
+CREATE STATISTICS ab1_exprstat_3 ON (a+b), a FROM ab1;
+SELECT stxkind FROM pg_statistic_ext WHERE stxname = 'ab1_exprstat_3';
+ stxkind
+-----------
+ {d,f,m,e}
+(1 row)
+
+-- date_trunc on timestamptz is not immutable, but that should not matter
+CREATE STATISTICS ab1_exprstat_4 ON date_trunc('day', d) FROM ab1;
+-- date_trunc on timestamp is immutable
+CREATE STATISTICS ab1_exprstat_5 ON date_trunc('day', c) FROM ab1;
+-- insert some data and run analyze, to test that these cases build properly
+INSERT INTO ab1
+SELECT
+ generate_series(1,10),
+ generate_series(1,10),
+ generate_series('2020-10-01'::timestamp, '2020-10-10'::timestamp, interval '1 day'),
+ generate_series('2020-10-01'::timestamptz, '2020-10-10'::timestamptz, interval '1 day');
+ANALYZE ab1;
+DROP TABLE ab1;
-- Verify supported object types for extended statistics
CREATE schema tststats;
CREATE TABLE tststats.t (a int, b int, c text);
@@ -427,6 +473,40 @@ SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE
1 | 1
(1 row)
+-- a => b, a => c, b => c
+TRUNCATE functional_dependencies;
+DROP STATISTICS func_deps_stat;
+-- now do the same thing, but with expressions
+INSERT INTO functional_dependencies (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+ANALYZE functional_dependencies;
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 35
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1 AND mod(c, 7) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 5
+(1 row)
+
+-- create statistics
+CREATE STATISTICS func_deps_stat (dependencies) ON (mod(a,11)), (mod(b::int, 13)), (mod(c, 7)) FROM functional_dependencies;
+ANALYZE functional_dependencies;
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1');
+ estimated | actual
+-----------+--------
+ 35 | 35
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1 AND mod(c, 7) = 1');
+ estimated | actual
+-----------+--------
+ 5 | 5
+(1 row)
+
-- a => b, a => c, b => c
TRUNCATE functional_dependencies;
DROP STATISTICS func_deps_stat;
@@ -896,6 +976,39 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 AND b =
1 | 1
(1 row)
+TRUNCATE mcv_lists;
+DROP STATISTICS mcv_lists_stats;
+-- random data (no MCV list), but with expression
+INSERT INTO mcv_lists (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+ANALYZE mcv_lists;
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 4
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1 AND mod(c,47) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 1
+(1 row)
+
+-- create statistics
+CREATE STATISTICS mcv_lists_stats (mcv) ON (mod(a,37)), (mod(b::int,41)), (mod(c,47)) FROM mcv_lists;
+ANALYZE mcv_lists;
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1');
+ estimated | actual
+-----------+--------
+ 3 | 4
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1 AND mod(c,47) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 1
+(1 row)
+
-- 100 distinct combinations, all in the MCV list
TRUNCATE mcv_lists;
DROP STATISTICS mcv_lists_stats;
@@ -1121,6 +1234,12 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 OR b = '
200 | 200
(1 row)
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 OR b = ''1'' OR c = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a IN (1, 2, 51, 52) AND b IN ( ''1'', ''2'')');
estimated | actual
-----------+--------
@@ -1207,6 +1326,458 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 AND b =
50 | 50
(1 row)
+-- 100 distinct combinations, all in the MCV list, but with expressions
+TRUNCATE mcv_lists;
+DROP STATISTICS mcv_lists_stats;
+INSERT INTO mcv_lists (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+ANALYZE mcv_lists;
+-- without any stats on the expressions, we have to use default selectivities, which
+-- is why the estimates here are different from the pre-computed case above
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+ estimated | actual
+-----------+--------
+ 556 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 556 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+ estimated | actual
+-----------+--------
+ 556 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 556 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+ estimated | actual
+-----------+--------
+ 185 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+ estimated | actual
+-----------+--------
+ 185 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+ estimated | actual
+-----------+--------
+ 185 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+ estimated | actual
+-----------+--------
+ 185 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 75 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 75 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+ estimated | actual
+-----------+--------
+ 1 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+ estimated | actual
+-----------+--------
+ 1 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+ estimated | actual
+-----------+--------
+ 1 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+ estimated | actual
+-----------+--------
+ 1 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+ estimated | actual
+-----------+--------
+ 53 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+ estimated | actual
+-----------+--------
+ 53 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 391 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+ estimated | actual
+-----------+--------
+ 391 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 6 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+ estimated | actual
+-----------+--------
+ 6 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 75 | 200
+(1 row)
+
+-- create statistics with expressions only (we create three separate stats, in order not to build more complex extended stats)
+CREATE STATISTICS mcv_lists_stats_1 ON (mod(a,100)) FROM mcv_lists;
+CREATE STATISTICS mcv_lists_stats_2 ON (mod(b::int,50)) FROM mcv_lists;
+CREATE STATISTICS mcv_lists_stats_3 ON (mod(c,25)) FROM mcv_lists;
+ANALYZE mcv_lists;
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+ estimated | actual
+-----------+--------
+ 1 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 343 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 343 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+ estimated | actual
+-----------+--------
+ 8 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+ estimated | actual
+-----------+--------
+ 8 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+ estimated | actual
+-----------+--------
+ 8 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+ estimated | actual
+-----------+--------
+ 8 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+ estimated | actual
+-----------+--------
+ 26 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+ estimated | actual
+-----------+--------
+ 26 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 10 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+ estimated | actual
+-----------+--------
+ 10 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 1 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+ estimated | actual
+-----------+--------
+ 1 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 343 | 200
+(1 row)
+
+DROP STATISTICS mcv_lists_stats_1;
+DROP STATISTICS mcv_lists_stats_2;
+DROP STATISTICS mcv_lists_stats_3;
+-- create statistics with both MCV and expressions
+CREATE STATISTICS mcv_lists_stats (mcv) ON (mod(a,100)), (mod(b::int,50)), (mod(c,25)) FROM mcv_lists;
+ANALYZE mcv_lists;
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+ estimated | actual
+-----------+--------
+ 50 | 50
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+ estimated | actual
+-----------+--------
+ 150 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+ estimated | actual
+-----------+--------
+ 150 | 150
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 100 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+ estimated | actual
+-----------+--------
+ 100 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+ estimated | actual
+-----------+--------
+ 100 | 100
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+ estimated | actual
+-----------+--------
+ 100 | 100
+(1 row)
+
+-- we can't use the statistic for OR clauses that are not fully covered (missing 'd' attribute)
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+ estimated | actual
+-----------+--------
+ 200 | 200
+(1 row)
+
-- 100 distinct combinations with NULL values, all in the MCV list
TRUNCATE mcv_lists;
DROP STATISTICS mcv_lists_stats;
@@ -1712,6 +2283,100 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists_multi WHERE a = 0 OR
(1 row)
DROP TABLE mcv_lists_multi;
+-- statistics on integer expressions
+CREATE TABLE expr_stats (a int, b int, c int);
+INSERT INTO expr_stats SELECT mod(i,10), mod(i,10), mod(i,10) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (2*a) = 0 AND (3*b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (a+b) = 0 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 1000
+(1 row)
+
+CREATE STATISTICS expr_stats_1 (mcv) ON (a+b), (a-b), (2*a), (3*b) FROM expr_stats;
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (2*a) = 0 AND (3*b) = 0');
+ estimated | actual
+-----------+--------
+ 1000 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (a+b) = 0 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1000 | 1000
+(1 row)
+
+DROP STATISTICS expr_stats_1;
+DROP TABLE expr_stats;
+-- statistics on a mix columns and expressions
+CREATE TABLE expr_stats (a int, b int, c int);
+INSERT INTO expr_stats SELECT mod(i,10), mod(i,10), mod(i,10) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (2*a) = 0 AND (3*b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 3 AND b = 3 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND b = 1 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 0
+(1 row)
+
+CREATE STATISTICS expr_stats_1 (mcv) ON a, b, (2*a), (3*b), (a+b), (a-b) FROM expr_stats;
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (2*a) = 0 AND (3*b) = 0');
+ estimated | actual
+-----------+--------
+ 1000 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 3 AND b = 3 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1000 | 1000
+(1 row)
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND b = 1 AND (a-b) = 0');
+ estimated | actual
+-----------+--------
+ 1 | 0
+(1 row)
+
+DROP TABLE expr_stats;
+-- statistics on expressions with different data types
+CREATE TABLE expr_stats (a int, b name, c text);
+INSERT INTO expr_stats SELECT mod(i,10), md5(mod(i,10)::text), md5(mod(i,10)::text) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (b || c) <= ''z'' AND (c || b) >= ''0''');
+ estimated | actual
+-----------+--------
+ 111 | 1000
+(1 row)
+
+CREATE STATISTICS expr_stats_1 (mcv) ON a, b, (b || c), (c || b) FROM expr_stats;
+ANALYZE expr_stats;
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (b || c) <= ''z'' AND (c || b) >= ''0''');
+ estimated | actual
+-----------+--------
+ 1000 | 1000
+(1 row)
+
+DROP TABLE expr_stats;
-- Permission tests. Users should not be able to see specific data values in
-- the extended statistics, if they lack permission to see those values in
-- the underlying table.
diff --git a/src/test/regress/sql/create_table_like.sql b/src/test/regress/sql/create_table_like.sql
index 06b76f949d..4929d373a2 100644
--- a/src/test/regress/sql/create_table_like.sql
+++ b/src/test/regress/sql/create_table_like.sql
@@ -124,7 +124,9 @@ CREATE TABLE ctlt1 (a text CHECK (length(a) > 2) PRIMARY KEY, b text);
CREATE INDEX ctlt1_b_key ON ctlt1 (b);
CREATE INDEX ctlt1_fnidx ON ctlt1 ((a || b));
CREATE STATISTICS ctlt1_a_b_stat ON a,b FROM ctlt1;
+CREATE STATISTICS ctlt1_expr_stat ON (a || b) FROM ctlt1;
COMMENT ON STATISTICS ctlt1_a_b_stat IS 'ab stats';
+COMMENT ON STATISTICS ctlt1_expr_stat IS 'ab expr stats';
COMMENT ON COLUMN ctlt1.a IS 'A';
COMMENT ON COLUMN ctlt1.b IS 'B';
COMMENT ON CONSTRAINT ctlt1_a_check ON ctlt1 IS 't1_a_check';
diff --git a/src/test/regress/sql/stats_ext.sql b/src/test/regress/sql/stats_ext.sql
index 0d7a114b19..bd2ada1676 100644
--- a/src/test/regress/sql/stats_ext.sql
+++ b/src/test/regress/sql/stats_ext.sql
@@ -28,16 +28,21 @@ end;
$$;
-- Verify failures
-CREATE TABLE ext_stats_test (x int, y int, z int);
+CREATE TABLE ext_stats_test (x text, y int, z int);
CREATE STATISTICS tst;
CREATE STATISTICS tst ON a, b;
CREATE STATISTICS tst FROM sometab;
CREATE STATISTICS tst ON a, b FROM nonexistent;
CREATE STATISTICS tst ON a, b FROM ext_stats_test;
CREATE STATISTICS tst ON x, x, y FROM ext_stats_test;
-CREATE STATISTICS tst ON x + y FROM ext_stats_test;
-CREATE STATISTICS tst ON (x, y) FROM ext_stats_test;
+CREATE STATISTICS tst ON x, x, y, x, x, y, x, x, y FROM ext_stats_test;
+CREATE STATISTICS tst ON x, x, y, x, x, (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1) FROM ext_stats_test;
+CREATE STATISTICS tst ON (x || 'x'), (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1), (x || 'x'), (x || 'x'), (y + 1) FROM ext_stats_test;
+CREATE STATISTICS tst ON (x || 'x'), (x || 'x'), y FROM ext_stats_test;
CREATE STATISTICS tst (unrecognized) ON x, y FROM ext_stats_test;
+-- incorrect expressions
+CREATE STATISTICS tst ON y + z FROM ext_stats_test; -- missing parentheses
+CREATE STATISTICS tst ON (x, y) FROM ext_stats_test; -- tuple expression
DROP TABLE ext_stats_test;
-- Ensure stats are dropped sanely, and test IF NOT EXISTS while at it
@@ -97,6 +102,36 @@ CREATE STATISTICS ab1_a_b_stats ON a, b FROM ab1;
ANALYZE ab1;
DROP TABLE ab1 CASCADE;
+-- basic test for statistics on expressions
+CREATE TABLE ab1 (a INTEGER, b INTEGER, c TIMESTAMP, d TIMESTAMPTZ);
+
+-- expression stats may be built on a single expression column
+CREATE STATISTICS ab1_exprstat_1 ON (a+b) FROM ab1;
+
+-- with a single expression, we only enable expression statistics
+CREATE STATISTICS ab1_exprstat_2 ON (a+b) FROM ab1;
+SELECT stxkind FROM pg_statistic_ext WHERE stxname = 'ab1_exprstat_2';
+
+-- adding anything to the expression builds all statistics kinds
+CREATE STATISTICS ab1_exprstat_3 ON (a+b), a FROM ab1;
+SELECT stxkind FROM pg_statistic_ext WHERE stxname = 'ab1_exprstat_3';
+
+-- date_trunc on timestamptz is not immutable, but that should not matter
+CREATE STATISTICS ab1_exprstat_4 ON date_trunc('day', d) FROM ab1;
+
+-- date_trunc on timestamp is immutable
+CREATE STATISTICS ab1_exprstat_5 ON date_trunc('day', c) FROM ab1;
+
+-- insert some data and run analyze, to test that these cases build properly
+INSERT INTO ab1
+SELECT
+ generate_series(1,10),
+ generate_series(1,10),
+ generate_series('2020-10-01'::timestamp, '2020-10-10'::timestamp, interval '1 day'),
+ generate_series('2020-10-01'::timestamptz, '2020-10-10'::timestamptz, interval '1 day');
+ANALYZE ab1;
+DROP TABLE ab1;
+
-- Verify supported object types for extended statistics
CREATE schema tststats;
@@ -272,6 +307,29 @@ SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE
TRUNCATE functional_dependencies;
DROP STATISTICS func_deps_stat;
+-- now do the same thing, but with expressions
+INSERT INTO functional_dependencies (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+
+ANALYZE functional_dependencies;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1 AND mod(c, 7) = 1');
+
+-- create statistics
+CREATE STATISTICS func_deps_stat (dependencies) ON (mod(a,11)), (mod(b::int, 13)), (mod(c, 7)) FROM functional_dependencies;
+
+ANALYZE functional_dependencies;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM functional_dependencies WHERE mod(a, 11) = 1 AND mod(b::int, 13) = 1 AND mod(c, 7) = 1');
+
+-- a => b, a => c, b => c
+TRUNCATE functional_dependencies;
+DROP STATISTICS func_deps_stat;
+
INSERT INTO functional_dependencies (a, b, c, filler1)
SELECT mod(i,100), mod(i,50), mod(i,25), i FROM generate_series(1,5000) s(i);
@@ -479,6 +537,28 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 AND b =
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 AND b = ''1'' AND c = 1');
+TRUNCATE mcv_lists;
+DROP STATISTICS mcv_lists_stats;
+
+-- random data (no MCV list), but with expression
+INSERT INTO mcv_lists (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+
+ANALYZE mcv_lists;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1 AND mod(c,47) = 1');
+
+-- create statistics
+CREATE STATISTICS mcv_lists_stats (mcv) ON (mod(a,37)), (mod(b::int,41)), (mod(c,47)) FROM mcv_lists;
+
+ANALYZE mcv_lists;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,37) = 1 AND mod(b::int,41) = 1 AND mod(c,47) = 1');
+
-- 100 distinct combinations, all in the MCV list
TRUNCATE mcv_lists;
DROP STATISTICS mcv_lists_stats;
@@ -565,6 +645,8 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 OR b = '
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 OR b = ''1'' OR c = 1 OR d IS NOT NULL');
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 OR b = ''1'' OR c = 1 OR d IS NOT NULL');
+
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a IN (1, 2, 51, 52) AND b IN ( ''1'', ''2'')');
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a IN (1, 2, 51, 52, NULL) AND b IN ( ''1'', ''2'', NULL)');
@@ -602,6 +684,180 @@ ANALYZE mcv_lists;
SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE a = 1 AND b = ''1''');
+
+-- 100 distinct combinations, all in the MCV list, but with expressions
+TRUNCATE mcv_lists;
+DROP STATISTICS mcv_lists_stats;
+
+INSERT INTO mcv_lists (a, b, c, filler1)
+ SELECT i, i, i, i FROM generate_series(1,5000) s(i);
+
+ANALYZE mcv_lists;
+
+-- without any stats on the expressions, we have to use default selectivities, which
+-- is why the estimates here are different from the pre-computed case above
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
+-- create statistics with expressions only (we create three separate stats, in order not to build more complex extended stats)
+CREATE STATISTICS mcv_lists_stats_1 ON (mod(a,100)) FROM mcv_lists;
+CREATE STATISTICS mcv_lists_stats_2 ON (mod(b::int,50)) FROM mcv_lists;
+CREATE STATISTICS mcv_lists_stats_3 ON (mod(c,25)) FROM mcv_lists;
+
+ANALYZE mcv_lists;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
+DROP STATISTICS mcv_lists_stats_1;
+DROP STATISTICS mcv_lists_stats_2;
+DROP STATISTICS mcv_lists_stats_3;
+
+-- create statistics with both MCV and expressions
+CREATE STATISTICS mcv_lists_stats (mcv) ON (mod(a,100)), (mod(b::int,50)), (mod(c,25)) FROM mcv_lists;
+
+ANALYZE mcv_lists;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 = mod(a,100) AND 1 = mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 1 AND mod(b::int,50) < 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 1 > mod(a,100) AND 1 > mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 0 AND mod(b::int,50) <= 0');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 0 >= mod(a,100) AND 0 >= mod(b::int,50)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 AND mod(b::int,50) = 1 AND mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND mod(b::int,50) < 1 AND mod(c,25) < 5');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < 5 AND 1 > mod(b::int,50) AND 5 > mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= 4 AND mod(b::int,50) <= 0 AND mod(c,25) <= 4');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE 4 >= mod(a,100) AND 0 >= mod(b::int,50) AND 4 >= mod(c,25)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52) AND mod(b::int,50) IN ( 1, 2)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) IN (1, 2, 51, 52, NULL) AND mod(b::int,50) IN ( 1, 2, NULL)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = ANY (ARRAY[NULL, 1, 2, 51, 52]) AND mod(b::int,50) = ANY (ARRAY[1, 2, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, 2, 3]) AND mod(b::int,50) IN (1, 2, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) <= ANY (ARRAY[1, NULL, 2, 3]) AND mod(b::int,50) IN (1, 2, NULL, 3)');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(c,25) > ANY (ARRAY[1, 2, 3, NULL])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, 3) AND mod(c,25) > ANY (ARRAY[1, 2, 3])');
+
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) < ALL (ARRAY[4, 5]) AND mod(b::int,50) IN (1, 2, NULL, 3) AND mod(c,25) > ANY (ARRAY[1, 2, NULL, 3])');
+
+-- we can't use the statistic for OR clauses that are not fully covered (missing 'd' attribute)
+SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists WHERE mod(a,100) = 1 OR mod(b::int,50) = 1 OR mod(c,25) = 1 OR d IS NOT NULL');
+
-- 100 distinct combinations with NULL values, all in the MCV list
TRUNCATE mcv_lists;
DROP STATISTICS mcv_lists_stats;
@@ -894,6 +1150,57 @@ SELECT * FROM check_estimated_rows('SELECT * FROM mcv_lists_multi WHERE a = 0 OR
DROP TABLE mcv_lists_multi;
+
+-- statistics on integer expressions
+CREATE TABLE expr_stats (a int, b int, c int);
+INSERT INTO expr_stats SELECT mod(i,10), mod(i,10), mod(i,10) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (2*a) = 0 AND (3*b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (a+b) = 0 AND (a-b) = 0');
+
+CREATE STATISTICS expr_stats_1 (mcv) ON (a+b), (a-b), (2*a), (3*b) FROM expr_stats;
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (2*a) = 0 AND (3*b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE (a+b) = 0 AND (a-b) = 0');
+
+DROP STATISTICS expr_stats_1;
+DROP TABLE expr_stats;
+
+-- statistics on a mix columns and expressions
+CREATE TABLE expr_stats (a int, b int, c int);
+INSERT INTO expr_stats SELECT mod(i,10), mod(i,10), mod(i,10) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (2*a) = 0 AND (3*b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 3 AND b = 3 AND (a-b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND b = 1 AND (a-b) = 0');
+
+CREATE STATISTICS expr_stats_1 (mcv) ON a, b, (2*a), (3*b), (a+b), (a-b) FROM expr_stats;
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (2*a) = 0 AND (3*b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 3 AND b = 3 AND (a-b) = 0');
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND b = 1 AND (a-b) = 0');
+
+DROP TABLE expr_stats;
+
+-- statistics on expressions with different data types
+CREATE TABLE expr_stats (a int, b name, c text);
+INSERT INTO expr_stats SELECT mod(i,10), md5(mod(i,10)::text), md5(mod(i,10)::text) FROM generate_series(1,10000) s(i);
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (b || c) <= ''z'' AND (c || b) >= ''0''');
+
+CREATE STATISTICS expr_stats_1 (mcv) ON a, b, (b || c), (c || b) FROM expr_stats;
+ANALYZE expr_stats;
+
+SELECT * FROM check_estimated_rows('SELECT * FROM expr_stats WHERE a = 0 AND (b || c) <= ''z'' AND (c || b) >= ''0''');
+
+DROP TABLE expr_stats;
+
+
-- Permission tests. Users should not be able to see specific data values in
-- the extended statistics, if they lack permission to see those values in
-- the underlying table.
--
2.26.2
--------------1F4A19FEB5D90C72707AFA7B--
^ permalink raw reply [nested|flat] 26+ messages in thread
* Memoize ANTI and SEMI JOIN inner
@ 2025-03-06 13:08 Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-06 13:08 UTC (permalink / raw)
To: pgsql-hackers
Hi,
In case of NestLoop with parameterised inner semi-join for each outer
tuple requires only a single tuple from its inner relation to produce a
result. It seems that the same principle applies to an anti-join. This
approach could effectively allow the Memoize node to enhance the
performance of pulled-up EXISTS and NOT EXISTS sublinks.
In attachment see a sketch of the feature. Since we are using single_row
mode, adapting this method to cache semi-join inner results should not
be extremely complex. However, I am unsure about potential corner cases
and would appreciate any feedback or criticisms regarding this approach.
--
regards, Andrei Lepikhov
From e47abde1f6a2c8c8d7fa588df72e59c0cd049d4c Mon Sep 17 00:00:00 2001
From: "Andrei V. Lepikhov" <[email protected]>
Date: Wed, 25 Dec 2024 14:33:12 +0700
Subject: [PATCH v0] Memoise the inner of SEMI- and ANTI-join.
To get the result of semi-join or anti-join, we only need one tuple from
the parameterised inner. The Postgres core already includes Memoize's
single_row mode for the case when it is proved that inner returns only a single
value. Thus, implementing similar logic for semi-joins and anti-joins is
quite feasible.
It is clear that the semi-join operates correctly: it returns the outer tuple
as soon as the first tuple is received from the inner query. In contrast, the
behaviour of the anti-join is less straightforward. If anti-join had a join
clause not pushed to the parameterised inner, it would be possible to read more
than a single tuple from the inner, which causes inconsistency. However,
it seems that this is an impossible case for the current planner.
---
src/backend/commands/explain.c | 9 +++++++++
src/backend/optimizer/path/costsize.c | 2 +-
src/backend/optimizer/path/joinpath.c | 5 +++--
src/backend/optimizer/util/pathnode.c | 2 +-
src/test/regress/expected/join.out | 3 ++-
src/test/regress/expected/subselect.out | 3 ++-
6 files changed, 18 insertions(+), 6 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index d8a7232cedb..c90067dadcb 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -3632,6 +3632,10 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
{
ExplainPropertyText("Cache Key", keystr.data, es);
ExplainPropertyText("Cache Mode", mstate->binary_mode ? "binary" : "logical", es);
+
+ /* Report only in the single mode case to not break current tests */
+ if (mstate->singlerow)
+ ExplainPropertyText("Store Mode", "singlerow", es);
}
else
{
@@ -3639,6 +3643,11 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
appendStringInfo(es->str, "Cache Key: %s\n", keystr.data);
ExplainIndentText(es);
appendStringInfo(es->str, "Cache Mode: %s\n", mstate->binary_mode ? "binary" : "logical");
+ if (mstate->singlerow)
+ {
+ ExplainIndentText(es);
+ appendStringInfo(es->str, "Store Mode: %s\n", "singlerow");
+ }
}
pfree(keystr.data);
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index 73d78617009..4cca5a7395c 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2545,7 +2545,7 @@ cost_memoize_rescan(PlannerInfo *root, MemoizePath *mpath,
ListCell *lc;
Cost input_startup_cost = mpath->subpath->startup_cost;
Cost input_total_cost = mpath->subpath->total_cost;
- double tuples = mpath->subpath->rows;
+ double tuples = mpath->singlerow ? 1 : mpath->subpath->rows;
double calls = mpath->calls;
int width = mpath->subpath->pathtarget->width;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..89738b483ef 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -682,6 +682,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ListCell *lc;
bool binary_mode;
List *ph_lateral_vars;
+ bool single_mode = false;
/* Obviously not if it's disabled */
if (!enable_memoize)
@@ -725,7 +726,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
*/
if (!extra->inner_unique && (jointype == JOIN_SEMI ||
jointype == JOIN_ANTI))
- return NULL;
+ single_mode = true;
/*
* Memoize normally marks cache entries as complete when it runs out of
@@ -808,7 +809,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || single_mode,
binary_mode,
outer_path->rows);
}
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 93e73cb44db..9b4c0f96193 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -1707,7 +1707,7 @@ create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
*/
pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
- pathnode->path.rows = subpath->rows;
+ pathnode->path.rows = (pathnode->singlerow) ? 1 : subpath->rows;
return pathnode;
}
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index a57bb18c24f..e8293c3c87d 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6469,10 +6469,11 @@ select * from sj t1
-> Memoize
Cache Key: t1.a, t1.b
Cache Mode: binary
+ Store Mode: singlerow
-> Sample Scan on sj
Sampling: system (t1.b)
Filter: (t1.a = a)
-(8 rows)
+(9 rows)
-- Ensure that SJE does not form a self-referential lateral dependency
explain (costs off)
diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out
index d0db8a412ff..5ab56f8d71f 100644
--- a/src/test/regress/expected/subselect.out
+++ b/src/test/regress/expected/subselect.out
@@ -2642,6 +2642,7 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Memoize
Cache Key: b.hundred, b.odd
Cache Mode: binary
+ Store Mode: singlerow
-> Subquery Scan on "ANY_subquery"
Filter: (b.hundred = "ANY_subquery".min)
-> Result
@@ -2650,5 +2651,5 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Index Scan using tenk2_hundred on tenk2 c
Index Cond: (hundred IS NOT NULL)
Filter: (odd = b.odd)
-(16 rows)
+(17 rows)
--
2.48.1
Attachments:
[text/plain] v0-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch (6.1K, ../../[email protected]/2-v0-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch)
download | inline diff:
From e47abde1f6a2c8c8d7fa588df72e59c0cd049d4c Mon Sep 17 00:00:00 2001
From: "Andrei V. Lepikhov" <[email protected]>
Date: Wed, 25 Dec 2024 14:33:12 +0700
Subject: [PATCH v0] Memoise the inner of SEMI- and ANTI-join.
To get the result of semi-join or anti-join, we only need one tuple from
the parameterised inner. The Postgres core already includes Memoize's
single_row mode for the case when it is proved that inner returns only a single
value. Thus, implementing similar logic for semi-joins and anti-joins is
quite feasible.
It is clear that the semi-join operates correctly: it returns the outer tuple
as soon as the first tuple is received from the inner query. In contrast, the
behaviour of the anti-join is less straightforward. If anti-join had a join
clause not pushed to the parameterised inner, it would be possible to read more
than a single tuple from the inner, which causes inconsistency. However,
it seems that this is an impossible case for the current planner.
---
src/backend/commands/explain.c | 9 +++++++++
src/backend/optimizer/path/costsize.c | 2 +-
src/backend/optimizer/path/joinpath.c | 5 +++--
src/backend/optimizer/util/pathnode.c | 2 +-
src/test/regress/expected/join.out | 3 ++-
src/test/regress/expected/subselect.out | 3 ++-
6 files changed, 18 insertions(+), 6 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index d8a7232cedb..c90067dadcb 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -3632,6 +3632,10 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
{
ExplainPropertyText("Cache Key", keystr.data, es);
ExplainPropertyText("Cache Mode", mstate->binary_mode ? "binary" : "logical", es);
+
+ /* Report only in the single mode case to not break current tests */
+ if (mstate->singlerow)
+ ExplainPropertyText("Store Mode", "singlerow", es);
}
else
{
@@ -3639,6 +3643,11 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
appendStringInfo(es->str, "Cache Key: %s\n", keystr.data);
ExplainIndentText(es);
appendStringInfo(es->str, "Cache Mode: %s\n", mstate->binary_mode ? "binary" : "logical");
+ if (mstate->singlerow)
+ {
+ ExplainIndentText(es);
+ appendStringInfo(es->str, "Store Mode: %s\n", "singlerow");
+ }
}
pfree(keystr.data);
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index 73d78617009..4cca5a7395c 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2545,7 +2545,7 @@ cost_memoize_rescan(PlannerInfo *root, MemoizePath *mpath,
ListCell *lc;
Cost input_startup_cost = mpath->subpath->startup_cost;
Cost input_total_cost = mpath->subpath->total_cost;
- double tuples = mpath->subpath->rows;
+ double tuples = mpath->singlerow ? 1 : mpath->subpath->rows;
double calls = mpath->calls;
int width = mpath->subpath->pathtarget->width;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..89738b483ef 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -682,6 +682,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ListCell *lc;
bool binary_mode;
List *ph_lateral_vars;
+ bool single_mode = false;
/* Obviously not if it's disabled */
if (!enable_memoize)
@@ -725,7 +726,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
*/
if (!extra->inner_unique && (jointype == JOIN_SEMI ||
jointype == JOIN_ANTI))
- return NULL;
+ single_mode = true;
/*
* Memoize normally marks cache entries as complete when it runs out of
@@ -808,7 +809,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || single_mode,
binary_mode,
outer_path->rows);
}
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 93e73cb44db..9b4c0f96193 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -1707,7 +1707,7 @@ create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
*/
pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
- pathnode->path.rows = subpath->rows;
+ pathnode->path.rows = (pathnode->singlerow) ? 1 : subpath->rows;
return pathnode;
}
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index a57bb18c24f..e8293c3c87d 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6469,10 +6469,11 @@ select * from sj t1
-> Memoize
Cache Key: t1.a, t1.b
Cache Mode: binary
+ Store Mode: singlerow
-> Sample Scan on sj
Sampling: system (t1.b)
Filter: (t1.a = a)
-(8 rows)
+(9 rows)
-- Ensure that SJE does not form a self-referential lateral dependency
explain (costs off)
diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out
index d0db8a412ff..5ab56f8d71f 100644
--- a/src/test/regress/expected/subselect.out
+++ b/src/test/regress/expected/subselect.out
@@ -2642,6 +2642,7 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Memoize
Cache Key: b.hundred, b.odd
Cache Mode: binary
+ Store Mode: singlerow
-> Subquery Scan on "ANY_subquery"
Filter: (b.hundred = "ANY_subquery".min)
-> Result
@@ -2650,5 +2651,5 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Index Scan using tenk2_hundred on tenk2 c
Index Cond: (hundred IS NOT NULL)
Filter: (odd = b.odd)
-(16 rows)
+(17 rows)
--
2.48.1
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
@ 2025-03-19 17:15 ` Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-19 17:15 UTC (permalink / raw)
To: pgsql-hackers
On 6/3/2025 14:08, Andrei Lepikhov wrote:
> Hi,
>
> In case of NestLoop with parameterised inner semi-join for each outer
> tuple requires only a single tuple from its inner relation to produce a
> result. It seems that the same principle applies to an anti-join. This
> approach could effectively allow the Memoize node to enhance the
> performance of pulled-up EXISTS and NOT EXISTS sublinks.
>
> In attachment see a sketch of the feature. Since we are using single_row
> mode, adapting this method to cache semi-join inner results should not
> be extremely complex. However, I am unsure about potential corner cases
> and would appreciate any feedback or criticisms regarding this approach.
I found a corner case that breaks this approach: when NestLoop has a
join clause or a filter, it may filter inner tuples, thus scanning more
than only one time for each outer. You can see the reproduction script
in the attachment. Each of the reproduction queries throws the error:
ERROR: cache entry already complete
How can we be sure that semi or anti-join needs only one tuple? I think
it would be enough to control the absence of join clauses and filters in
the join. Unfortunately, we only have such a guarantee in the plan
creation stage (maybe even setrefs.c). So, it seems we need to invent an
approach like AlternativeSubplan.
--
regards, Andrei Lepikhov
Attachments:
[application/sql] repro.sql (1.3K, ../../[email protected]/2-repro.sql)
download
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
@ 2025-03-20 06:02 ` David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
0 siblings, 3 replies; 26+ messages in thread
From: David Rowley @ 2025-03-20 06:02 UTC (permalink / raw)
To: Andrei Lepikhov <[email protected]>; +Cc: pgsql-hackers
On Thu, 20 Mar 2025 at 06:16, Andrei Lepikhov <[email protected]> wrote:
> How can we be sure that semi or anti-join needs only one tuple? I think
> it would be enough to control the absence of join clauses and filters in
> the join. Unfortunately, we only have such a guarantee in the plan
> creation stage (maybe even setrefs.c). So, it seems we need to invent an
> approach like AlternativeSubplan.
I suggest looking at what 9e215378d did. You might be able to also
allow semi and anti-joins providing the cache keys cover the entire
join condition. I think this might be safe as Nested Loop will only
ask its inner subnode for the first match before skipping to the next
outer row and with anti-join, there's no reason to look for additional
rows after the first. Semi-join and unique joins do the same thing in
nodeNestloop.c. To save doing additional checks at run-time, the code
does:
nlstate->js.single_match = (node->join.inner_unique ||
node->join.jointype == JOIN_SEMI);
For making this work, I think the attached should be about the guts of
the code changes. I didn't look at the comments. Right now I can't
think of any reason why this can't be done, but some experimentation
might reveal some reason that it can't.
David
Attachments:
[application/octet-stream] memoize_semi_and_anti_joins_experiment.patch (1.7K, ../../CAApHDvr2NRbjLXEUXvvKNCDcHfhC6UBL6ZmJzXZ29MBPi3+y8g@mail.gmail.com/2-memoize_semi_and_anti_joins_experiment.patch)
download | inline diff:
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..03b4506e873 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -714,19 +714,6 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ph_lateral_vars == NIL)
return NULL;
- /*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
- *
- * XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
- * = true. Should we? See add_paths_to_joinrel()
- */
- if (!extra->inner_unique && (jointype == JOIN_SEMI ||
- jointype == JOIN_ANTI))
- return NULL;
-
/*
* Memoize normally marks cache entries as complete when it runs out of
* tuples to read from its subplan. However, with unique joins, Nested
@@ -753,7 +740,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
* the inner scan's filter instead of the join filter. Maybe it's worth
* considering doing that?
*/
- if (extra->inner_unique &&
+ if ((extra->inner_unique || jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
(inner_path->param_info == NULL ||
bms_num_members(inner_path->param_info->ppi_serials) <
list_length(extra->restrictlist)))
@@ -808,7 +795,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || jointype == JOIN_SEMI || jointype == JOIN_ANTI,
binary_mode,
outer_path->rows);
}
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-21 15:56 ` Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2 siblings, 1 reply; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-21 15:56 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: pgsql-hackers
On 20/3/2025 07:02, David Rowley wrote:
> On Thu, 20 Mar 2025 at 06:16, Andrei Lepikhov <[email protected]> wrote:
>> How can we be sure that semi or anti-join needs only one tuple? I think
>> it would be enough to control the absence of join clauses and filters in
>> the join. Unfortunately, we only have such a guarantee in the plan
>> creation stage (maybe even setrefs.c). So, it seems we need to invent an
>> approach like AlternativeSubplan.
>
> I suggest looking at what 9e215378d did. You might be able to also
> allow semi and anti-joins providing the cache keys cover the entire
> join condition. I think this might be safe as Nested Loop will only
> ask its inner subnode for the first match before skipping to the next
> outer row and with anti-join, there's no reason to look for additional
> rows after the first. Semi-join and unique joins do the same thing in
> nodeNestloop.c. To save doing additional checks at run-time, the code
> does:
Thank you for the clue! I almost took the wrong direction.
I have attached the new patch, which includes corrected comments for
better clarification of the changes, as well as some additional tests.
I now feel much more confident about this version since I have resolved
that concern.
--
regards, Andrei Lepikhov
From c5897e31d2a95de04fa0e641aeff43f3118bcced Mon Sep 17 00:00:00 2001
From: "Andrei V. Lepikhov" <[email protected]>
Date: Fri, 21 Mar 2025 15:55:40 +0100
Subject: [PATCH v1] Memoise the inner of SEMI- and ANTI-join.
To produce the result of semi-join or anti-join, we only need one tuple from
the parameterised inner. The Postgres core already includes Memoize's
single_row mode for the case when it is proved that inner returns only a single
value. Thus, implementing similar logic for semi-joins and anti-joins is
quite doable.
Usually, these types of join need only single tuple from the inner to produce
result for each outer tuple. But if after pushing parameterised clauses down to
the inner the NestLoop still have some join clauses or filters, it may reject
some inner tuples during execution and call the inner more than once. To prevent
that we check that all the restrictions have been pushed to the inner.
---
src/backend/commands/explain.c | 4 +
src/backend/optimizer/path/costsize.c | 2 +-
src/backend/optimizer/path/joinpath.c | 21 +++--
src/backend/optimizer/util/pathnode.c | 2 +-
src/test/regress/expected/join.out | 3 +-
src/test/regress/expected/memoize.out | 102 ++++++++++++++++++++++++
src/test/regress/expected/subselect.out | 3 +-
src/test/regress/sql/memoize.sql | 49 ++++++++++++
8 files changed, 174 insertions(+), 12 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index 391b34a2af2..1b4b3b740ab 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -3628,6 +3628,10 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
ExplainPropertyText("Cache Key", keystr.data, es);
ExplainPropertyText("Cache Mode", mstate->binary_mode ? "binary" : "logical", es);
+ /* Report only in the single mode case to not break current tests */
+ if (mstate->singlerow)
+ ExplainPropertyText("Store Mode", "singlerow", es);
+
pfree(keystr.data);
if (!es->analyze)
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index f6f77b8fe19..941ba6e1d49 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2545,7 +2545,7 @@ cost_memoize_rescan(PlannerInfo *root, MemoizePath *mpath,
ListCell *lc;
Cost input_startup_cost = mpath->subpath->startup_cost;
Cost input_total_cost = mpath->subpath->total_cost;
- double tuples = mpath->subpath->rows;
+ double tuples = mpath->singlerow ? 1 : mpath->subpath->rows;
double calls = mpath->calls;
int width = mpath->subpath->pathtarget->width;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..c75408f552b 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -682,6 +682,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ListCell *lc;
bool binary_mode;
List *ph_lateral_vars;
+ bool single_mode = false;
/* Obviously not if it's disabled */
if (!enable_memoize)
@@ -715,23 +716,27 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
+ * We may do this for SEMI or ANTI joins when they need only one tuple from
+ * the inner side to produce the result. Following if condition checks that
+ * rule.
*
* XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
* = true. Should we? See add_paths_to_joinrel()
*/
if (!extra->inner_unique && (jointype == JOIN_SEMI ||
jointype == JOIN_ANTI))
- return NULL;
+ single_mode = true;
/*
* Memoize normally marks cache entries as complete when it runs out of
* tuples to read from its subplan. However, with unique joins, Nested
* Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. This means that we may not read the inner side of the
+ * inner tuple. Another case is a semi or anti join. If number of join
+ * clauses, pushed to the inner as parameterised filter no less than the
+ * number of join clauses, that means all the clauses have been pushed to
+ * the inner and any tuple coming from the inner side will be successfully
+ * used to build the join result.
+ * This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
* automatically mark cache entries as complete after fetching the first
@@ -753,7 +758,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
* the inner scan's filter instead of the join filter. Maybe it's worth
* considering doing that?
*/
- if (extra->inner_unique &&
+ if ((extra->inner_unique || single_mode) &&
(inner_path->param_info == NULL ||
bms_num_members(inner_path->param_info->ppi_serials) <
list_length(extra->restrictlist)))
@@ -808,7 +813,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || single_mode,
binary_mode,
outer_path->rows);
}
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 93e73cb44db..9b4c0f96193 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -1707,7 +1707,7 @@ create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
*/
pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
- pathnode->path.rows = subpath->rows;
+ pathnode->path.rows = (pathnode->singlerow) ? 1 : subpath->rows;
return pathnode;
}
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index a57bb18c24f..e8293c3c87d 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6469,10 +6469,11 @@ select * from sj t1
-> Memoize
Cache Key: t1.a, t1.b
Cache Mode: binary
+ Store Mode: singlerow
-> Sample Scan on sj
Sampling: system (t1.b)
Filter: (t1.a = a)
-(8 rows)
+(9 rows)
-- Ensure that SJE does not form a self-referential lateral dependency
explain (costs off)
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..3ad473b211d 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -500,3 +500,105 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ Join Filter: ((c.x = a.x) AND (c.z = a.z))
+ -> Nested Loop
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = b.x)
+ Filter: (y = b.y)
+(13 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ Join Filter: (c.y = b.y)
+ -> Nested Loop Left Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: ((x = a.x) AND (z = a.z))
+(12 rows)
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out
index d0db8a412ff..5ab56f8d71f 100644
--- a/src/test/regress/expected/subselect.out
+++ b/src/test/regress/expected/subselect.out
@@ -2642,6 +2642,7 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Memoize
Cache Key: b.hundred, b.odd
Cache Mode: binary
+ Store Mode: singlerow
-> Subquery Scan on "ANY_subquery"
Filter: (b.hundred = "ANY_subquery".min)
-> Result
@@ -2650,5 +2651,5 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Index Scan using tenk2_hundred on tenk2 c
Index Cond: (hundred IS NOT NULL)
Filter: (odd = b.odd)
-(16 rows)
+(17 rows)
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..048c8e90ad0 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -244,3 +244,52 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
--
2.48.1
Attachments:
[text/plain] v1-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch (14.2K, ../../[email protected]/2-v1-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch)
download | inline diff:
From c5897e31d2a95de04fa0e641aeff43f3118bcced Mon Sep 17 00:00:00 2001
From: "Andrei V. Lepikhov" <[email protected]>
Date: Fri, 21 Mar 2025 15:55:40 +0100
Subject: [PATCH v1] Memoise the inner of SEMI- and ANTI-join.
To produce the result of semi-join or anti-join, we only need one tuple from
the parameterised inner. The Postgres core already includes Memoize's
single_row mode for the case when it is proved that inner returns only a single
value. Thus, implementing similar logic for semi-joins and anti-joins is
quite doable.
Usually, these types of join need only single tuple from the inner to produce
result for each outer tuple. But if after pushing parameterised clauses down to
the inner the NestLoop still have some join clauses or filters, it may reject
some inner tuples during execution and call the inner more than once. To prevent
that we check that all the restrictions have been pushed to the inner.
---
src/backend/commands/explain.c | 4 +
src/backend/optimizer/path/costsize.c | 2 +-
src/backend/optimizer/path/joinpath.c | 21 +++--
src/backend/optimizer/util/pathnode.c | 2 +-
src/test/regress/expected/join.out | 3 +-
src/test/regress/expected/memoize.out | 102 ++++++++++++++++++++++++
src/test/regress/expected/subselect.out | 3 +-
src/test/regress/sql/memoize.sql | 49 ++++++++++++
8 files changed, 174 insertions(+), 12 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index 391b34a2af2..1b4b3b740ab 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -3628,6 +3628,10 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
ExplainPropertyText("Cache Key", keystr.data, es);
ExplainPropertyText("Cache Mode", mstate->binary_mode ? "binary" : "logical", es);
+ /* Report only in the single mode case to not break current tests */
+ if (mstate->singlerow)
+ ExplainPropertyText("Store Mode", "singlerow", es);
+
pfree(keystr.data);
if (!es->analyze)
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index f6f77b8fe19..941ba6e1d49 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2545,7 +2545,7 @@ cost_memoize_rescan(PlannerInfo *root, MemoizePath *mpath,
ListCell *lc;
Cost input_startup_cost = mpath->subpath->startup_cost;
Cost input_total_cost = mpath->subpath->total_cost;
- double tuples = mpath->subpath->rows;
+ double tuples = mpath->singlerow ? 1 : mpath->subpath->rows;
double calls = mpath->calls;
int width = mpath->subpath->pathtarget->width;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..c75408f552b 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -682,6 +682,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ListCell *lc;
bool binary_mode;
List *ph_lateral_vars;
+ bool single_mode = false;
/* Obviously not if it's disabled */
if (!enable_memoize)
@@ -715,23 +716,27 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
+ * We may do this for SEMI or ANTI joins when they need only one tuple from
+ * the inner side to produce the result. Following if condition checks that
+ * rule.
*
* XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
* = true. Should we? See add_paths_to_joinrel()
*/
if (!extra->inner_unique && (jointype == JOIN_SEMI ||
jointype == JOIN_ANTI))
- return NULL;
+ single_mode = true;
/*
* Memoize normally marks cache entries as complete when it runs out of
* tuples to read from its subplan. However, with unique joins, Nested
* Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. This means that we may not read the inner side of the
+ * inner tuple. Another case is a semi or anti join. If number of join
+ * clauses, pushed to the inner as parameterised filter no less than the
+ * number of join clauses, that means all the clauses have been pushed to
+ * the inner and any tuple coming from the inner side will be successfully
+ * used to build the join result.
+ * This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
* automatically mark cache entries as complete after fetching the first
@@ -753,7 +758,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
* the inner scan's filter instead of the join filter. Maybe it's worth
* considering doing that?
*/
- if (extra->inner_unique &&
+ if ((extra->inner_unique || single_mode) &&
(inner_path->param_info == NULL ||
bms_num_members(inner_path->param_info->ppi_serials) <
list_length(extra->restrictlist)))
@@ -808,7 +813,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || single_mode,
binary_mode,
outer_path->rows);
}
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 93e73cb44db..9b4c0f96193 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -1707,7 +1707,7 @@ create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
*/
pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
- pathnode->path.rows = subpath->rows;
+ pathnode->path.rows = (pathnode->singlerow) ? 1 : subpath->rows;
return pathnode;
}
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index a57bb18c24f..e8293c3c87d 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6469,10 +6469,11 @@ select * from sj t1
-> Memoize
Cache Key: t1.a, t1.b
Cache Mode: binary
+ Store Mode: singlerow
-> Sample Scan on sj
Sampling: system (t1.b)
Filter: (t1.a = a)
-(8 rows)
+(9 rows)
-- Ensure that SJE does not form a self-referential lateral dependency
explain (costs off)
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..3ad473b211d 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -500,3 +500,105 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ Join Filter: ((c.x = a.x) AND (c.z = a.z))
+ -> Nested Loop
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = b.x)
+ Filter: (y = b.y)
+(13 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ Join Filter: (c.y = b.y)
+ -> Nested Loop Left Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: ((x = a.x) AND (z = a.z))
+(12 rows)
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out
index d0db8a412ff..5ab56f8d71f 100644
--- a/src/test/regress/expected/subselect.out
+++ b/src/test/regress/expected/subselect.out
@@ -2642,6 +2642,7 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Memoize
Cache Key: b.hundred, b.odd
Cache Mode: binary
+ Store Mode: singlerow
-> Subquery Scan on "ANY_subquery"
Filter: (b.hundred = "ANY_subquery".min)
-> Result
@@ -2650,5 +2651,5 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Index Scan using tenk2_hundred on tenk2 c
Index Cond: (hundred IS NOT NULL)
Filter: (odd = b.odd)
-(16 rows)
+(17 rows)
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..048c8e90ad0 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -244,3 +244,52 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
--
2.48.1
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
@ 2025-03-31 02:33 ` Alena Rybakina <[email protected]>
2025-03-31 02:50 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
0 siblings, 2 replies; 26+ messages in thread
From: Alena Rybakina @ 2025-03-31 02:33 UTC (permalink / raw)
To: Andrei Lepikhov <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
Hi!
On 21.03.2025 18:56, Andrei Lepikhov wrote:
> On 20/3/2025 07:02, David Rowley wrote:
>> On Thu, 20 Mar 2025 at 06:16, Andrei Lepikhov <[email protected]> wrote:
>>> How can we be sure that semi or anti-join needs only one tuple? I think
>>> it would be enough to control the absence of join clauses and
>>> filters in
>>> the join. Unfortunately, we only have such a guarantee in the plan
>>> creation stage (maybe even setrefs.c). So, it seems we need to
>>> invent an
>>> approach like AlternativeSubplan.
>>
>> I suggest looking at what 9e215378d did. You might be able to also
>> allow semi and anti-joins providing the cache keys cover the entire
>> join condition. I think this might be safe as Nested Loop will only
>> ask its inner subnode for the first match before skipping to the next
>> outer row and with anti-join, there's no reason to look for additional
>> rows after the first. Semi-join and unique joins do the same thing in
>> nodeNestloop.c. To save doing additional checks at run-time, the code
>> does:
> Thank you for the clue! I almost took the wrong direction.
> I have attached the new patch, which includes corrected comments for
> better clarification of the changes, as well as some additional tests.
> I now feel much more confident about this version since I have
> resolved that concern.
>
I reviewed your patch and made a couple of suggestions.
The first change is related to your comment (and the one before it). I
fixed some grammar issues and simplified the wording to make it clearer
and easier to understand.
The second change involves adding an Assert when generating the Memoize
path. Based on the existing comment and the surrounding logic (shown
below),
I believe it's worth asserting that both inner_unique and single_mode
are not true at the same time — just as a safety check.
/*
* We may do this for SEMI or ANTI joins when they need only one tuple from
* the inner side to produce the result. Following if condition checks that
* rule.
*
* XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
* = true. Should we? See add_paths_to_joinrel()
*/
if(!extra->inner_unique&& (jointype== JOIN_SEMI||
jointype== JOIN_ANTI))
single_mode= true;
--
Regards,
Alena Rybakina
Postgres Professional
Attachments:
[text/x-patch] memoize.diff (2.0K, ../../[email protected]/3-memoize.diff)
download | inline diff:
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index c75408f552b..252cb712943 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -728,14 +728,16 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
single_mode = true;
/*
- * Memoize normally marks cache entries as complete when it runs out of
- * tuples to read from its subplan. However, with unique joins, Nested
- * Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. Another case is a semi or anti join. If number of join
- * clauses, pushed to the inner as parameterised filter no less than the
- * number of join clauses, that means all the clauses have been pushed to
- * the inner and any tuple coming from the inner side will be successfully
- * used to build the join result.
+ * Normally, memoize marks cache entries as complete when it exhausts
+ * all tuples from its subplan. However, in unique joins, Nested Loop
+ * will skip to the next outer tuple after finding the first matching
+ * inner tuple.
+ * Another case is a SEMI or ANTI joins. If the number of join clauses,
+ * pushed to the inner as parameterised filter is equal to or greater
+ * than the total number of join clauses. This implies that all relevant
+ * join conditions have been applied on the inner side, so any returned
+ * inner tuple will be guaranteed to satisfy the join condition, making
+ * it safe to memoize.
* This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
@@ -808,6 +810,8 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
&hash_operators,
&binary_mode))
{
+ Assert(!(extra->inner_unique && single_mode));
+
return (Path *) create_memoize_path(root,
innerrel,
inner_path,
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
@ 2025-03-31 02:50 ` Alena Rybakina <[email protected]>
1 sibling, 0 replies; 26+ messages in thread
From: Alena Rybakina @ 2025-03-31 02:50 UTC (permalink / raw)
To: Andrei Lepikhov <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
I realized that I uploaded my diff file with a small mistake - sorry
about that. I've corrected it with this message so your tests can pass
in the CI.
On 31.03.2025 05:33, Alena Rybakina wrote:
>
> Hi!
>
> On 21.03.2025 18:56, Andrei Lepikhov wrote:
>> On 20/3/2025 07:02, David Rowley wrote:
>>> On Thu, 20 Mar 2025 at 06:16, Andrei Lepikhov <[email protected]>
>>> wrote:
>>>> How can we be sure that semi or anti-join needs only one tuple? I
>>>> think
>>>> it would be enough to control the absence of join clauses and
>>>> filters in
>>>> the join. Unfortunately, we only have such a guarantee in the plan
>>>> creation stage (maybe even setrefs.c). So, it seems we need to
>>>> invent an
>>>> approach like AlternativeSubplan.
>>>
>>> I suggest looking at what 9e215378d did. You might be able to also
>>> allow semi and anti-joins providing the cache keys cover the entire
>>> join condition. I think this might be safe as Nested Loop will only
>>> ask its inner subnode for the first match before skipping to the next
>>> outer row and with anti-join, there's no reason to look for additional
>>> rows after the first. Semi-join and unique joins do the same thing in
>>> nodeNestloop.c. To save doing additional checks at run-time, the code
>>> does:
>> Thank you for the clue! I almost took the wrong direction.
>> I have attached the new patch, which includes corrected comments for
>> better clarification of the changes, as well as some additional tests.
>> I now feel much more confident about this version since I have
>> resolved that concern.
>>
>
> I reviewed your patch and made a couple of suggestions.
>
> The first change is related to your comment (and the one before it). I
> fixed some grammar issues and simplified the wording to make it
> clearer and easier to understand.
>
> The second change involves adding an Assert when generating the
> Memoize path. Based on the existing comment and the surrounding logic
> (shown below),
> I believe it's worth asserting that both inner_unique and single_mode
> are not true at the same time — just as a safety check.
> /*
> * We may do this for SEMI or ANTI joins when they need only one tuple from
> * the inner side to produce the result. Following if condition checks that
> * rule.
> *
> * XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
> * = true. Should we? See add_paths_to_joinrel()
> */
> if(!extra->inner_unique&& (jointype== JOIN_SEMI||
> jointype== JOIN_ANTI))
> single_mode= true;
--
Regards,
Alena Rybakina
Postgres Professional
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index c75408f552b..252cb712943 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -728,14 +728,16 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
single_mode = true;
/*
- * Memoize normally marks cache entries as complete when it runs out of
- * tuples to read from its subplan. However, with unique joins, Nested
- * Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. Another case is a semi or anti join. If number of join
- * clauses, pushed to the inner as parameterised filter no less than the
- * number of join clauses, that means all the clauses have been pushed to
- * the inner and any tuple coming from the inner side will be successfully
- * used to build the join result.
+ * Normally, memoize marks cache entries as complete when it exhausts
+ * all tuples from its subplan. However, in unique joins, Nested Loop
+ * will skip to the next outer tuple after finding the first matching
+ * inner tuple.
+ * Another case is a SEMI or ANTI joins. If the number of join clauses,
+ * pushed to the inner as parameterised filter is equal to or greater
+ * than the total number of join clauses. This implies that all relevant
+ * join conditions have been applied on the inner side, so any returned
+ * inner tuple will be guaranteed to satisfy the join condition, making
+ * it safe to memoize.
* This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
@@ -808,6 +810,8 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
&hash_operators,
&binary_mode))
{
+ Assert(!(extra->inner_unique && single_mode));
+
return (Path *) create_memoize_path(root,
innerrel,
inner_path,
Attachments:
[text/plain] memoize.diff.no-cfbot (2.0K, ../../[email protected]/3-memoize.diff.no-cfbot)
download | inline diff:
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index c75408f552b..252cb712943 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -728,14 +728,16 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
single_mode = true;
/*
- * Memoize normally marks cache entries as complete when it runs out of
- * tuples to read from its subplan. However, with unique joins, Nested
- * Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. Another case is a semi or anti join. If number of join
- * clauses, pushed to the inner as parameterised filter no less than the
- * number of join clauses, that means all the clauses have been pushed to
- * the inner and any tuple coming from the inner side will be successfully
- * used to build the join result.
+ * Normally, memoize marks cache entries as complete when it exhausts
+ * all tuples from its subplan. However, in unique joins, Nested Loop
+ * will skip to the next outer tuple after finding the first matching
+ * inner tuple.
+ * Another case is a SEMI or ANTI joins. If the number of join clauses,
+ * pushed to the inner as parameterised filter is equal to or greater
+ * than the total number of join clauses. This implies that all relevant
+ * join conditions have been applied on the inner side, so any returned
+ * inner tuple will be guaranteed to satisfy the join condition, making
+ * it safe to memoize.
* This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
@@ -808,6 +810,8 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
&hash_operators,
&binary_mode))
{
+ Assert(!(extra->inner_unique && single_mode));
+
return (Path *) create_memoize_path(root,
innerrel,
inner_path,
[text/x-patch] v1-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch (14.2K, ../../[email protected]/4-v1-0001-Memoise-the-inner-of-SEMI-and-ANTI-join.patch)
download | inline diff:
From c5897e31d2a95de04fa0e641aeff43f3118bcced Mon Sep 17 00:00:00 2001
From: "Andrei V. Lepikhov" <[email protected]>
Date: Fri, 21 Mar 2025 15:55:40 +0100
Subject: [PATCH v1] Memoise the inner of SEMI- and ANTI-join.
To produce the result of semi-join or anti-join, we only need one tuple from
the parameterised inner. The Postgres core already includes Memoize's
single_row mode for the case when it is proved that inner returns only a single
value. Thus, implementing similar logic for semi-joins and anti-joins is
quite doable.
Usually, these types of join need only single tuple from the inner to produce
result for each outer tuple. But if after pushing parameterised clauses down to
the inner the NestLoop still have some join clauses or filters, it may reject
some inner tuples during execution and call the inner more than once. To prevent
that we check that all the restrictions have been pushed to the inner.
---
src/backend/commands/explain.c | 4 +
src/backend/optimizer/path/costsize.c | 2 +-
src/backend/optimizer/path/joinpath.c | 21 +++--
src/backend/optimizer/util/pathnode.c | 2 +-
src/test/regress/expected/join.out | 3 +-
src/test/regress/expected/memoize.out | 102 ++++++++++++++++++++++++
src/test/regress/expected/subselect.out | 3 +-
src/test/regress/sql/memoize.sql | 49 ++++++++++++
8 files changed, 174 insertions(+), 12 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index 391b34a2af2..1b4b3b740ab 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -3628,6 +3628,10 @@ show_memoize_info(MemoizeState *mstate, List *ancestors, ExplainState *es)
ExplainPropertyText("Cache Key", keystr.data, es);
ExplainPropertyText("Cache Mode", mstate->binary_mode ? "binary" : "logical", es);
+ /* Report only in the single mode case to not break current tests */
+ if (mstate->singlerow)
+ ExplainPropertyText("Store Mode", "singlerow", es);
+
pfree(keystr.data);
if (!es->analyze)
diff --git a/src/backend/optimizer/path/costsize.c b/src/backend/optimizer/path/costsize.c
index f6f77b8fe19..941ba6e1d49 100644
--- a/src/backend/optimizer/path/costsize.c
+++ b/src/backend/optimizer/path/costsize.c
@@ -2545,7 +2545,7 @@ cost_memoize_rescan(PlannerInfo *root, MemoizePath *mpath,
ListCell *lc;
Cost input_startup_cost = mpath->subpath->startup_cost;
Cost input_total_cost = mpath->subpath->total_cost;
- double tuples = mpath->subpath->rows;
+ double tuples = mpath->singlerow ? 1 : mpath->subpath->rows;
double calls = mpath->calls;
int width = mpath->subpath->pathtarget->width;
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..c75408f552b 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -682,6 +682,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
ListCell *lc;
bool binary_mode;
List *ph_lateral_vars;
+ bool single_mode = false;
/* Obviously not if it's disabled */
if (!enable_memoize)
@@ -715,23 +716,27 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
+ * We may do this for SEMI or ANTI joins when they need only one tuple from
+ * the inner side to produce the result. Following if condition checks that
+ * rule.
*
* XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
* = true. Should we? See add_paths_to_joinrel()
*/
if (!extra->inner_unique && (jointype == JOIN_SEMI ||
jointype == JOIN_ANTI))
- return NULL;
+ single_mode = true;
/*
* Memoize normally marks cache entries as complete when it runs out of
* tuples to read from its subplan. However, with unique joins, Nested
* Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. This means that we may not read the inner side of the
+ * inner tuple. Another case is a semi or anti join. If number of join
+ * clauses, pushed to the inner as parameterised filter no less than the
+ * number of join clauses, that means all the clauses have been pushed to
+ * the inner and any tuple coming from the inner side will be successfully
+ * used to build the join result.
+ * This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
* automatically mark cache entries as complete after fetching the first
@@ -753,7 +758,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
* the inner scan's filter instead of the join filter. Maybe it's worth
* considering doing that?
*/
- if (extra->inner_unique &&
+ if ((extra->inner_unique || single_mode) &&
(inner_path->param_info == NULL ||
bms_num_members(inner_path->param_info->ppi_serials) <
list_length(extra->restrictlist)))
@@ -808,7 +813,7 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
inner_path,
param_exprs,
hash_operators,
- extra->inner_unique,
+ extra->inner_unique || single_mode,
binary_mode,
outer_path->rows);
}
diff --git a/src/backend/optimizer/util/pathnode.c b/src/backend/optimizer/util/pathnode.c
index 93e73cb44db..9b4c0f96193 100644
--- a/src/backend/optimizer/util/pathnode.c
+++ b/src/backend/optimizer/util/pathnode.c
@@ -1707,7 +1707,7 @@ create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
*/
pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
- pathnode->path.rows = subpath->rows;
+ pathnode->path.rows = (pathnode->singlerow) ? 1 : subpath->rows;
return pathnode;
}
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index a57bb18c24f..e8293c3c87d 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6469,10 +6469,11 @@ select * from sj t1
-> Memoize
Cache Key: t1.a, t1.b
Cache Mode: binary
+ Store Mode: singlerow
-> Sample Scan on sj
Sampling: system (t1.b)
Filter: (t1.a = a)
-(8 rows)
+(9 rows)
-- Ensure that SJE does not form a self-referential lateral dependency
explain (costs off)
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..3ad473b211d 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -500,3 +500,105 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+ QUERY PLAN
+----------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ Store Mode: singlerow
+ -> Index Only Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = a.x)
+(9 rows)
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Semi Join
+ Join Filter: ((c.x = a.x) AND (c.z = a.z))
+ -> Nested Loop
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: (x = b.x)
+ Filter: (y = b.y)
+(13 rows)
+
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+ QUERY PLAN
+---------------------------------------------------------------------------------
+ Nested Loop Anti Join
+ Join Filter: (c.y = b.y)
+ -> Nested Loop Left Join
+ -> Seq Scan on mem_semi_inner_a a
+ Disabled: true
+ -> Memoize
+ Cache Key: a.x
+ Cache Mode: logical
+ -> Index Scan using mem_semi_inner_b_x_idx on mem_semi_inner_b b
+ Index Cond: (x = a.x)
+ -> Index Scan using mem_semi_inner_c_x_z_idx on mem_semi_inner_c c
+ Index Cond: ((x = a.x) AND (z = a.z))
+(12 rows)
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
diff --git a/src/test/regress/expected/subselect.out b/src/test/regress/expected/subselect.out
index d0db8a412ff..5ab56f8d71f 100644
--- a/src/test/regress/expected/subselect.out
+++ b/src/test/regress/expected/subselect.out
@@ -2642,6 +2642,7 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Memoize
Cache Key: b.hundred, b.odd
Cache Mode: binary
+ Store Mode: singlerow
-> Subquery Scan on "ANY_subquery"
Filter: (b.hundred = "ANY_subquery".min)
-> Result
@@ -2650,5 +2651,5 @@ ON B.hundred in (SELECT min(c.hundred) FROM tenk2 C WHERE c.odd = b.odd);
-> Index Scan using tenk2_hundred on tenk2 c
Index Cond: (hundred IS NOT NULL)
Filter: (odd = b.odd)
-(16 rows)
+(17 rows)
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..048c8e90ad0 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -244,3 +244,52 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+--
+-- Tests on Memoize under SEMI and ANTI joins.
+--
+
+CREATE TABLE mem_semi_inner_a (x int, z int);
+CREATE TABLE mem_semi_inner_b (x int, y int);
+CREATE TABLE mem_semi_inner_c (x int, y int, z int);
+INSERT INTO mem_semi_inner_a (x,z)
+ (SELECT value%2, -42 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_b (x,y)
+ (SELECT value%5, -value%5-1 FROM generate_series(1,10) AS value);
+INSERT INTO mem_semi_inner_c (x,y,z)
+ (SELECT value%50, value, -42 FROM generate_series(1,100) AS value);
+CREATE INDEX ON mem_semi_inner_b(x);
+CREATE INDEX ON mem_semi_inner_c(x,z);
+VACUUM ANALYZE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+
+-- Force NestLoop and IndexScan. Hope, the Memoize node win the cost
+-- competition on the inner c table scan.
+SET enable_hashjoin = 'off';
+SET enable_mergejoin = 'off';
+SET enable_seqscan = 'off';
+
+-- Primitive example of semi and anti join caching the inner's result
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+EXPLAIN (COSTS OFF)
+SELECT a.x FROM mem_semi_inner_a a
+WHERE NOT EXISTS (SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x);
+
+-- Check the query plans contain the Memoize node over the "Scan b" operator
+-- and does not contain memoize over "Scan c".
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+EXPLAIN (COSTS OFF)
+SELECT a.x, b.x FROM mem_semi_inner_a a
+ LEFT JOIN mem_semi_inner_b b ON (a.x=b.x)
+WHERE NOT EXISTS (
+ SELECT 1 FROM mem_semi_inner_c c WHERE c.x=a.x AND c.y=b.y AND c.z=a.z);
+
+DROP TABLE mem_semi_inner_a,mem_semi_inner_b,mem_semi_inner_c;
+RESET enable_hashjoin;
+RESET enable_mergejoin;
+RESET enable_seqscan;
--
2.48.1
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
@ 2025-03-31 03:04 ` David Rowley <[email protected]>
2025-03-31 03:21 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
1 sibling, 1 reply; 26+ messages in thread
From: David Rowley @ 2025-03-31 03:04 UTC (permalink / raw)
To: Alena Rybakina <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Mon, 31 Mar 2025 at 15:33, Alena Rybakina <[email protected]> wrote:
> I believe it's worth asserting that both inner_unique and single_mode are not true at the same time — just as a safety check.
add_paths_to_joinrel() just chooses not to populate inner_unique for
SEMI and ANTI joins because, as of today's master, it's pretty
pointless to determine that because the executor will short-circuit
and skip to the next outer tuple for those join types anyway. I don't
follow why having both these flags set would cause trouble. It seems
perfectly legitimate that add_paths_to_joinrel() could choose to set
the inner_unique flag for these join types, and if it did, the Assert
you're proposing would fail for no good reason.
David
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-31 03:21 ` Alena Rybakina <[email protected]>
2025-03-31 03:33 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: Alena Rybakina @ 2025-03-31 03:21 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On 31.03.2025 06:04, David Rowley wrote:
> On Mon, 31 Mar 2025 at 15:33, Alena Rybakina<[email protected]> wrote:
>> I believe it's worth asserting that both inner_unique and single_mode are not true at the same time — just as a safety check.
> add_paths_to_joinrel() just chooses not to populate inner_unique for
> SEMI and ANTI joins because, as of today's master, it's pretty
> pointless to determine that because the executor will short-circuit
> and skip to the next outer tuple for those join types anyway. I don't
> follow why having both these flags set would cause trouble. It seems
> perfectly legitimate that add_paths_to_joinrel() could choose to set
> the inner_unique flag for these join types, and if it did, the Assert
> you're proposing would fail for no good reason.
>
I tend to agree with you that someone might set this flag to true for
these join types in the future.
However, is it necessary to check that extra->inner_unique must be false
for SEMI/ANTI joins here, or am I missing something? It looks a little
confusing at this point.
if(!extra->inner_unique&& (jointype== JOIN_SEMI|| jointype== JOIN_ANTI))
single_mode= true;
--
Regards,
Alena Rybakina
Postgres Professional
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 03:21 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
@ 2025-03-31 03:33 ` David Rowley <[email protected]>
2025-03-31 06:34 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 07:45 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
0 siblings, 2 replies; 26+ messages in thread
From: David Rowley @ 2025-03-31 03:33 UTC (permalink / raw)
To: Alena Rybakina <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Mon, 31 Mar 2025 at 16:21, Alena Rybakina <[email protected]> wrote:
> However, is it necessary to check that extra->inner_unique must be false for SEMI/ANTI joins here, or am I missing something? It looks a little confusing at this point.
If it is necessary, I don't see the reason for it. It was me that
worked on unique joins and I see no reason why a SEMI or ANTI join
couldn't be marked as unique. The reason they're not today is that the
only point of the unique join optimisation is so that during
execution, the join nodes could skip to the next outer tuple after
matching the current outer to an inner. If the join is unique, then
there are no more join partners to find for the current outer after
matching it up once. With SEMI and ANTI joins, we skip to the next
outer tuple after finding a match anyway, so there's no point in going
to the trouble of setting the inner_unique flag.
I can't say definitively that we won't find a reason in the future
that we should set inner_unique for SEMI/ANTI joins, so I don't follow
the need for the Assert.
Maybe you're seeing something that I'm not. What do you think will
break if both flags are true?
David
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 03:21 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:33 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-31 06:34 ` Alena Rybakina <[email protected]>
1 sibling, 0 replies; 26+ messages in thread
From: Alena Rybakina @ 2025-03-31 06:34 UTC (permalink / raw)
To: David Rowley <[email protected]>; Andrei Lepikhov <[email protected]>; +Cc: pgsql-hackers
On 31.03.2025 06:33, David Rowley wrote:
> On Mon, 31 Mar 2025 at 16:21, Alena Rybakina <[email protected]> wrote:
>> However, is it necessary to check that extra->inner_unique must be false for SEMI/ANTI joins here, or am I missing something? It looks a little confusing at this point.
> If it is necessary, I don't see the reason for it. It was me that
> worked on unique joins and I see no reason why a SEMI or ANTI join
> couldn't be marked as unique. The reason they're not today is that the
> only point of the unique join optimisation is so that during
> execution, the join nodes could skip to the next outer tuple after
> matching the current outer to an inner. If the join is unique, then
> there are no more join partners to find for the current outer after
> matching it up once. With SEMI and ANTI joins, we skip to the next
> outer tuple after finding a match anyway, so there's no point in going
> to the trouble of setting the inner_unique flag.
>
> I can't say definitively that we won't find a reason in the future
> that we should set inner_unique for SEMI/ANTI joins, so I don't follow
> the need for the Assert.
>
> Maybe you're seeing something that I'm not. What do you think will
> break if both flags are true?
>
Actually, I was mainly confused by the code itself - the check seemed to
contradict the explanation. It looked like we were enforcing that
inner_unique must be false for SEMI/ANTI joins, even though it's not
actually important for those join types.
That’s why I originally proposed either adding an Assert or removing
this flag from check altogether, just to make it more explicit.
So, I agree with your explanation — by the definition of SEMI and ANTI
joins, there's no need to set inner_unique, and also no need to assert
against it.
These joins skip to the next outer tuple once they find a match (or fail
to find one, in the case of ANTI).
I updated the diff, where I left changes only in the code comment.
--
Regards,
Alena Rybakina
Postgres Professional
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index c75408f552b..252cb712943 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -728,14 +728,16 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
single_mode = true;
/*
- * Memoize normally marks cache entries as complete when it runs out of
- * tuples to read from its subplan. However, with unique joins, Nested
- * Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. Another case is a semi or anti join. If number of join
- * clauses, pushed to the inner as parameterised filter no less than the
- * number of join clauses, that means all the clauses have been pushed to
- * the inner and any tuple coming from the inner side will be successfully
- * used to build the join result.
+ * Normally, memoize marks cache entries as complete when it exhausts
+ * all tuples from its subplan. However, in unique joins, Nested Loop
+ * will skip to the next outer tuple after finding the first matching
+ * inner tuple.
+ * Another case is a SEMI or ANTI joins. If the number of join clauses,
+ * pushed to the inner as parameterised filter is equal to or greater
+ * than the total number of join clauses. This implies that all relevant
+ * join conditions have been applied on the inner side, so any returned
+ * inner tuple will be guaranteed to satisfy the join condition, making
+ * it safe to memoize.
* This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
Attachments:
[text/plain] memoize.diff.no-cfbot (1.7K, ../../[email protected]/2-memoize.diff.no-cfbot)
download | inline diff:
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index c75408f552b..252cb712943 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -728,14 +728,16 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
single_mode = true;
/*
- * Memoize normally marks cache entries as complete when it runs out of
- * tuples to read from its subplan. However, with unique joins, Nested
- * Loop will skip to the next outer tuple after finding the first matching
- * inner tuple. Another case is a semi or anti join. If number of join
- * clauses, pushed to the inner as parameterised filter no less than the
- * number of join clauses, that means all the clauses have been pushed to
- * the inner and any tuple coming from the inner side will be successfully
- * used to build the join result.
+ * Normally, memoize marks cache entries as complete when it exhausts
+ * all tuples from its subplan. However, in unique joins, Nested Loop
+ * will skip to the next outer tuple after finding the first matching
+ * inner tuple.
+ * Another case is a SEMI or ANTI joins. If the number of join clauses,
+ * pushed to the inner as parameterised filter is equal to or greater
+ * than the total number of join clauses. This implies that all relevant
+ * join conditions have been applied on the inner side, so any returned
+ * inner tuple will be guaranteed to satisfy the join condition, making
+ * it safe to memoize.
* This means that we may not read the inner side of the
* join to completion which leaves no opportunity to mark the cache entry
* as complete. To work around that, when the join is unique we
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 03:21 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:33 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-31 07:45 ` Andrei Lepikhov <[email protected]>
1 sibling, 0 replies; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-31 07:45 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: pgsql-hackers; Alena Rybakina <[email protected]>
On 3/31/25 05:33, David Rowley wrote:
> I can't say definitively that we won't find a reason in the future
> that we should set inner_unique for SEMI/ANTI joins, so I don't follow
> the need for the Assert.
>
> Maybe you're seeing something that I'm not. What do you think will
> break if both flags are true?
I considered targeting PG 19 and July Comitfest for this feature, but
currently, I don't see any further development needed here. What are
your thoughts, David? Does it make sense to commit this to PG 18?
I have no reason to rush the process, but this feature seems beneficial
for practice. When the internal structure is a bushy join tree,
producing even a single tuple can be costly. SQL Server's Spool node
addresses this issue, and people sometimes experience confusion
detecting degradation during migration with specific queries.
--
regards, Andrei Lepikhov
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-31 09:03 ` Richard Guo <[email protected]>
2025-03-31 09:39 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:46 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2 siblings, 2 replies; 26+ messages in thread
From: Richard Guo @ 2025-03-31 09:03 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Thu, Mar 20, 2025 at 3:02 PM David Rowley <[email protected]> wrote:
> For making this work, I think the attached should be about the guts of
> the code changes. I didn't look at the comments. Right now I can't
> think of any reason why this can't be done, but some experimentation
> might reveal some reason that it can't.
I reviewed this patch and I have some concerns about the following
code:
if (extra->inner_unique &&
(inner_path->param_info == NULL ||
bms_num_members(inner_path->param_info->ppi_serials) <
list_length(extra->restrictlist)))
return NULL;
I understand that this check is used to ensure that the entire join
condition is parameterized in the case of unique joins, so that we can
safely mark the cache entry as complete after reading the first tuple.
However, ppi_clauses includes join clauses available from all outer
rels, not just the current outer rel, while extra->restrictlist only
includes the restriction clauses for the current join. This means the
check could pass even if a restriction clause isn't parameterized, as
long as another join clause, which doesn't belong to the current join,
is included in ppi_clauses.
Thanks
Richard
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-03-31 09:39 ` David Rowley <[email protected]>
2025-03-31 09:50 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
1 sibling, 1 reply; 26+ messages in thread
From: David Rowley @ 2025-03-31 09:39 UTC (permalink / raw)
To: Richard Guo <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Mon, 31 Mar 2025 at 22:03, Richard Guo <[email protected]> wrote:
> I reviewed this patch and I have some concerns about the following
> code:
>
> if (extra->inner_unique &&
> (inner_path->param_info == NULL ||
> bms_num_members(inner_path->param_info->ppi_serials) <
> list_length(extra->restrictlist)))
> return NULL;
>
> I understand that this check is used to ensure that the entire join
> condition is parameterized in the case of unique joins, so that we can
> safely mark the cache entry as complete after reading the first tuple.
> However, ppi_clauses includes join clauses available from all outer
> rels, not just the current outer rel, while extra->restrictlist only
> includes the restriction clauses for the current join. This means the
> check could pass even if a restriction clause isn't parameterized, as
> long as another join clause, which doesn't belong to the current join,
> is included in ppi_clauses.
Shouldn't you be more concerned about master here than the patch given
that the code you pasted is from master?
David
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:39 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-03-31 09:50 ` Richard Guo <[email protected]>
0 siblings, 0 replies; 26+ messages in thread
From: Richard Guo @ 2025-03-31 09:50 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Mon, Mar 31, 2025 at 6:39 PM David Rowley <[email protected]> wrote:
> On Mon, 31 Mar 2025 at 22:03, Richard Guo <[email protected]> wrote:
> > I reviewed this patch and I have some concerns about the following
> > code:
> >
> > if (extra->inner_unique &&
> > (inner_path->param_info == NULL ||
> > bms_num_members(inner_path->param_info->ppi_serials) <
> > list_length(extra->restrictlist)))
> > return NULL;
> >
> > I understand that this check is used to ensure that the entire join
> > condition is parameterized in the case of unique joins, so that we can
> > safely mark the cache entry as complete after reading the first tuple.
> > However, ppi_clauses includes join clauses available from all outer
> > rels, not just the current outer rel, while extra->restrictlist only
> > includes the restriction clauses for the current join. This means the
> > check could pass even if a restriction clause isn't parameterized, as
> > long as another join clause, which doesn't belong to the current join,
> > is included in ppi_clauses.
>
> Shouldn't you be more concerned about master here than the patch given
> that the code you pasted is from master?
Right. This code is from the master branch, not the patch. It caught
my attention while I was reviewing the patch and noticed it modifies
this code.
Thanks
Richard
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-03-31 09:46 ` Andrei Lepikhov <[email protected]>
2025-03-31 10:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
1 sibling, 1 reply; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-31 09:46 UTC (permalink / raw)
To: Richard Guo <[email protected]>; David Rowley <[email protected]>; +Cc: pgsql-hackers
On 3/31/25 11:03, Richard Guo wrote:
> On Thu, Mar 20, 2025 at 3:02 PM David Rowley <[email protected]> wrote:
>> For making this work, I think the attached should be about the guts of
>> the code changes. I didn't look at the comments. Right now I can't
>> think of any reason why this can't be done, but some experimentation
>> might reveal some reason that it can't.
>
> I reviewed this patch and I have some concerns about the following
> code:
>
> if (extra->inner_unique &&
> (inner_path->param_info == NULL ||
> bms_num_members(inner_path->param_info->ppi_serials) <
> list_length(extra->restrictlist)))
> return NULL;
>
> I understand that this check is used to ensure that the entire join
> condition is parameterized in the case of unique joins, so that we can
> safely mark the cache entry as complete after reading the first tuple.
> However, ppi_clauses includes join clauses available from all outer
> rels, not just the current outer rel, while extra->restrictlist only
> includes the restriction clauses for the current join. This means the
> check could pass even if a restriction clause isn't parameterized, as
> long as another join clause, which doesn't belong to the current join,
> is included in ppi_clauses.
Initially, I had the same concern. But if ppi_clauses contains a qual,
it should refer to this join and, as a result, be in the
extra->restrictlist, isn't it?
I thought about references to the other side of an OUTER JOIN, but if
the subquery refers to LHS, it just not be transformed to the SEMI/ANTI
join.
Anyway, if you provide an example or just a sketch, I will be happy to
discover it.
--
regards, Andrei Lepikhov
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:46 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
@ 2025-03-31 10:18 ` Richard Guo <[email protected]>
2025-03-31 10:33 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: Richard Guo @ 2025-03-31 10:18 UTC (permalink / raw)
To: Andrei Lepikhov <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
On Mon, Mar 31, 2025 at 6:46 PM Andrei Lepikhov <[email protected]> wrote:
> On 3/31/25 11:03, Richard Guo wrote:
> > I reviewed this patch and I have some concerns about the following
> > code:
> >
> > if (extra->inner_unique &&
> > (inner_path->param_info == NULL ||
> > bms_num_members(inner_path->param_info->ppi_serials) <
> > list_length(extra->restrictlist)))
> > return NULL;
> >
> > I understand that this check is used to ensure that the entire join
> > condition is parameterized in the case of unique joins, so that we can
> > safely mark the cache entry as complete after reading the first tuple.
> > However, ppi_clauses includes join clauses available from all outer
> > rels, not just the current outer rel, while extra->restrictlist only
> > includes the restriction clauses for the current join. This means the
> > check could pass even if a restriction clause isn't parameterized, as
> > long as another join clause, which doesn't belong to the current join,
> > is included in ppi_clauses.
> Initially, I had the same concern. But if ppi_clauses contains a qual,
> it should refer to this join and, as a result, be in the
> extra->restrictlist, isn't it?
Hmm, I don't think so. As I mentioned upthread, ppi_clauses includes
join clauses available from all outer rels, not just the current one.
So a clause included in ppi_clauses is not necessarily included in
extra->restrictlist. As an example, consider
create table t (a int, b int);
explain (costs off)
select * from t t1 join t t2 join
lateral (select *, t1.a+t2.a as x from t t3 offset 0) t3
on t2.a = t3.a
on t1.b = t3.b;
QUERY PLAN
---------------------------------------------------------
Nested Loop
-> Seq Scan on t t2
-> Nested Loop
-> Seq Scan on t t1
-> Subquery Scan on t3
Filter: ((t2.a = t3.a) AND (t1.b = t3.b))
-> Seq Scan on t t3_1
(7 rows)
t3's ppi_clauses includes "t2.a = t3.a" and "t1.b = t3.b", while t1/t3
join's restrictlist only includes "t1.b = t3.b".
Thanks
Richard
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:46 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 10:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-03-31 10:33 ` Andrei Lepikhov <[email protected]>
2025-04-01 07:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: Andrei Lepikhov @ 2025-03-31 10:33 UTC (permalink / raw)
To: Richard Guo <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
On 3/31/25 12:18, Richard Guo wrote:
> On Mon, Mar 31, 2025 at 6:46 PM Andrei Lepikhov <[email protected]> wrote:
> Nested Loop
> -> Seq Scan on t t2
> -> Nested Loop
> -> Seq Scan on t t1
> -> Subquery Scan on t3
> Filter: ((t2.a = t3.a) AND (t1.b = t3.b))
> -> Seq Scan on t t3_1
> (7 rows)
>
> t3's ppi_clauses includes "t2.a = t3.a" and "t1.b = t3.b", while t1/t3
> join's restrictlist only includes "t1.b = t3.b".
I attempted to make your query ab it closer to our case:
SET enable_mergejoin = f;
SET enable_hashjoin = f;
SET enable_material = f;
CREATE INDEX ON t(a);
explain (costs off)
select * from t t1 join t t2
on EXISTS (select *, t1.a+t2.a as x from t t3
WHERE t2.a = t3.a AND t1.b = t3.b);
and I don't get the case. As I see, ANTI/SEMI join just transforms to
the regular join and it is still not the case. May you be more specific?
--
regards, Andrei Lepikhov
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:46 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 10:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 10:33 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
@ 2025-04-01 07:18 ` Richard Guo <[email protected]>
0 siblings, 0 replies; 26+ messages in thread
From: Richard Guo @ 2025-04-01 07:18 UTC (permalink / raw)
To: Andrei Lepikhov <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
On Mon, Mar 31, 2025 at 7:33 PM Andrei Lepikhov <[email protected]> wrote:
> and I don't get the case. As I see, ANTI/SEMI join just transforms to
> the regular join and it is still not the case. May you be more specific?
Upthread, you said that a qual contained in ppi_clauses will also be
included in extra->restrictlist. I provided this example to show that
this is not true. And it seems to me that this discrepancy makes the
check I mentioned earlier not reliable in all cases. As I explained
earlier, this check could pass even if a restriction clause isn't
parameterized, as long as another join clause, which doesn't belong to
the current join, is included in ppi_clauses. This is not correct.
This isn't about your patch, but about the master code. However, if
this code is incorrect, your patch will also behave incorrectly, since
you patch relies on and extends this check.
Thanks
Richard
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
@ 2025-04-09 06:48 ` Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2 siblings, 1 reply; 26+ messages in thread
From: Richard Guo @ 2025-04-09 06:48 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Thu, Mar 20, 2025 at 3:02 PM David Rowley <[email protected]> wrote:
> For making this work, I think the attached should be about the guts of
> the code changes. I didn't look at the comments. Right now I can't
> think of any reason why this can't be done, but some experimentation
> might reveal some reason that it can't.
I conducted some experiments, and I'm afraid it's not safe to consider
Memoize for semi or anti joins, unless the inner side is provably
unique. As an example, please consider:
create table t (a int, b boolean);
insert into t select i%3, false from generate_series(1,100)i;
analyze t;
select * from t t1 where t1.a in
(select a from t t2 where t2.b in
(select t1.b from t t3 where t2.a > 1 offset 0));
ERROR: cache entry already complete
With the proposed patch, this query results in an error.
The problem is that join clauses from the upper level may be moved to
the semi join. For a given outer tuple, the first inner tuple that
satisfies the current parameters will mark the cache entry as complete
because singlerow is set to true. However, if that inner tuple and
the current outer tuple don't satisfy the join clauses, the second
inner tuple that satisfies the parameters will complain that the cache
entry is already marked as complete.
If the inner side is provably unique, there will be no such problem,
as there won't be a second matching tuple. OTOH, in this case, the
semi join will be reduced to an inner join by reduce_unique_semijoins.
Therefore, it doesn't make much sense to prove inner_unique for semi
joins in add_paths_to_joinrel.
Perhaps we could spend some planner cycles proving inner_unique for
anti joins, so that Memoize nodes can be considered for them?
Thanks
Richard
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-04-10 10:36 ` Richard Guo <[email protected]>
2025-05-13 13:51 ` Re: Memoize ANTI and SEMI JOIN inner Andres Freund <[email protected]>
2025-07-02 02:08 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
0 siblings, 2 replies; 26+ messages in thread
From: Richard Guo @ 2025-04-10 10:36 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Wed, Apr 9, 2025 at 6:18 PM David Rowley <[email protected]> wrote:
> On Wed, 9 Apr 2025 at 18:48, Richard Guo <[email protected]> wrote:
> > Perhaps we could spend some planner cycles proving inner_unique for
> > anti joins, so that Memoize nodes can be considered for them?
> Worth a try. It should be pretty easy to enable, as far as I can see.
> It might just be a case of shuffling the cases around in the switch
> statement in add_paths_to_joinrel().
Right. Here is a patch for that.
Thanks
Richard
Attachments:
[application/octet-stream] v1-0001-Enable-use-of-Memoize-for-ANTI-joins.patch (9.1K, ../../CAMbWs48FdLiMNrmJL-g6mDvoQVt0yNyJAqMkv4e2Pk-5GKCZLA@mail.gmail.com/2-v1-0001-Enable-use-of-Memoize-for-ANTI-joins.patch)
download | inline diff:
From 942b88bdb6a2dd789e49fdcc947c748cac259a76 Mon Sep 17 00:00:00 2001
From: Richard Guo <[email protected]>
Date: Wed, 9 Apr 2025 18:06:48 +0900
Subject: [PATCH v1] Enable use of Memoize for ANTI joins
Currently, we do not support Memoize for SEMI and ANTI joins because
nested loop SEMI/ANTI joins do not scan the inner relation to
completion, which prevents Memoize from marking the cache entry as
complete. One might argue that we could mark the cache entry as
complete after fetching the first inner tuple, but that would not be
safe: if the first inner tuple and the current outer tuple do not
satisfy the join clauses, a second inner tuple matching the parameters
would find the cache entry already marked as complete.
However, if the inner side is provably unique, this issue doesn't
arise, since there would be no second matching tuple. That said, this
doesn't help in the case of SEMI joins, because a SEMI join with a
provably unique inner side would already have been reduced to an inner
join by reduce_unique_semijoins.
Therefore, in this patch, we check whether the inner relation is
provably unique for ANTI joins and enable the use of Memoize in such
cases.
---
src/backend/optimizer/path/joinpath.c | 47 +++++++++++----------
src/test/regress/expected/memoize.out | 59 +++++++++++++++++++++++++++
src/test/regress/sql/memoize.sql | 26 ++++++++++++
3 files changed, 110 insertions(+), 22 deletions(-)
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 18891ce9156..064e73cc862 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -154,13 +154,17 @@ add_paths_to_joinrel(PlannerInfo *root,
/*
* See if the inner relation is provably unique for this outer rel.
*
- * We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
- * matter since the executor can make the equivalent optimization anyway;
- * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, we
- * must be considering a semijoin whose inner side is not provably unique
- * (else reduce_unique_semijoins would've simplified it), so there's no
- * point in calling innerrel_is_unique. However, if the LHS covers all of
- * the semijoin's min_lefthand, then it's appropriate to set inner_unique
+ * We have some special cases: for JOIN_SEMI, it doesn't matter since the
+ * executor can make the equivalent optimization anyway. It also doesn't
+ * help enable use of Memoize, since a semijoin with a provably unique
+ * inner side should have been reduced to an inner join in that case.
+ * Therefore, we need not expend planner cycles on proofs. (For
+ * JOIN_ANTI, although it doesn't help the executor for the same reason,
+ * it can benefit Memoize paths.) For JOIN_UNIQUE_INNER, we must be
+ * considering a semijoin whose inner side is not provably unique (else
+ * reduce_unique_semijoins would've simplified it), so there's no point in
+ * calling innerrel_is_unique. However, if the LHS covers all of the
+ * semijoin's min_lefthand, then it's appropriate to set inner_unique
* because the path produced by create_unique_path will be unique relative
* to the LHS. (If we have an LHS that's only part of the min_lefthand,
* that is *not* true.) For JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid
@@ -169,12 +173,6 @@ add_paths_to_joinrel(PlannerInfo *root,
switch (jointype)
{
case JOIN_SEMI:
- case JOIN_ANTI:
-
- /*
- * XXX it may be worth proving this to allow a Memoize to be
- * considered for Nested Loop Semi/Anti Joins.
- */
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
@@ -715,16 +713,21 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
- *
- * XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
- * = true. Should we? See add_paths_to_joinrel()
+ * Currently we don't do this for SEMI and ANTI joins, because nested loop
+ * SEMI/ANTI joins don't scan the inner node to completion, which means
+ * memoize cannot mark the cache entry as complete. Nor can we mark the
+ * cache entry as complete after fetching the first inner tuple, because
+ * if that tuple and the current outer tuple don't satisfy the join
+ * clauses, a second inner tuple that satisfies the parameters would find
+ * the cache entry already marked as complete. The only exception is when
+ * the inner relation is provably unique, as in that case, there won't be
+ * a second matching tuple and we can safely mark the cache entry as
+ * complete after fetching the first inner tuple. Note that in such
+ * cases, the SEMI join should have been reduced to an inner join by
+ * reduce_unique_semijoins.
*/
- if (!extra->inner_unique && (jointype == JOIN_SEMI ||
- jointype == JOIN_ANTI))
+ if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
+ !extra->inner_unique)
return NULL;
/*
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..73ced02e65b 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -500,3 +500,62 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+ explain_memoize
+--------------------------------------------------------------------------------------------
+ Aggregate (actual rows=1.00 loops=N)
+ -> Nested Loop Anti Join (actual rows=33.00 loops=N)
+ -> Seq Scan on tab_anti t1 (actual rows=100.00 loops=N)
+ -> Memoize (actual rows=0.67 loops=N)
+ Cache Key: (t1.a + 1), t1.a
+ Cache Mode: binary
+ Hits: 97 Misses: 3 Evictions: Zero Overflows: 0 Memory Usage: NkB
+ -> Subquery Scan on t2 (actual rows=0.67 loops=N)
+ Filter: ((t1.a + 1) = t2.a)
+ Rows Removed by Filter: 2
+ -> Unique (actual rows=2.67 loops=N)
+ -> Sort (actual rows=67.33 loops=N)
+ Sort Key: t2_1.a
+ Sort Method: quicksort Memory: 27kB
+ -> Seq Scan on tab_anti t2_1 (actual rows=100.00 loops=N)
+(15 rows)
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+ count
+-------
+ 33
+(1 row)
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+ QUERY PLAN
+-------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on tab_anti t1
+ -> Nested Loop Semi Join
+ Join Filter: (t1.a = t2.a)
+ -> Seq Scan on tab_anti t2
+ -> Subquery Scan on "ANY_subquery"
+ Filter: (t2.b = "ANY_subquery".b)
+ -> Result
+ One-Time Filter: (t2.a > 1)
+ -> Seq Scan on tab_anti t3
+(10 rows)
+
+DROP TABLE tab_anti;
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..1370d2c9cfc 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -244,3 +244,29 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+
+DROP TABLE tab_anti;
--
2.43.0
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-05-13 13:51 ` Andres Freund <[email protected]>
2025-05-14 02:13 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
1 sibling, 1 reply; 26+ messages in thread
From: Andres Freund @ 2025-05-13 13:51 UTC (permalink / raw)
To: Richard Guo <[email protected]>; +Cc: David Rowley <[email protected]>; Andrei Lepikhov <[email protected]>; pgsql-hackers
Hi,
On 2025-04-10 19:36:14 +0900, Richard Guo wrote:
> On Wed, Apr 9, 2025 at 6:18 PM David Rowley <[email protected]> wrote:
> > On Wed, 9 Apr 2025 at 18:48, Richard Guo <[email protected]> wrote:
> > > Perhaps we could spend some planner cycles proving inner_unique for
> > > anti joins, so that Memoize nodes can be considered for them?
>
> > Worth a try. It should be pretty easy to enable, as far as I can see.
> > It might just be a case of shuffling the cases around in the switch
> > statement in add_paths_to_joinrel().
>
> Right. Here is a patch for that.
This is failing on CI:
https://cirrus-ci.com/github/postgresql-cfbot/postgresql/cf%2F5636
https://api.cirrus-ci.com/v1/artifact/task/5411026402803712/testrun/build-32/testrun/regress/regress...
diff -U3 /tmp/cirrus-ci-build/src/test/regress/expected/memoize.out /tmp/cirrus-ci-build/build-32/testrun/regress/regress/results/memoize.out
--- /tmp/cirrus-ci-build/src/test/regress/expected/memoize.out 2025-04-12 11:24:10.866868945 +0000
+++ /tmp/cirrus-ci-build/build-32/testrun/regress/regress/results/memoize.out 2025-04-12 11:32:44.454864476 +0000
@@ -525,7 +525,7 @@
-> Unique (actual rows=2.67 loops=N)
-> Sort (actual rows=67.33 loops=N)
Sort Key: t2_1.a
- Sort Method: quicksort Memory: 27kB
+ Sort Method: quicksort Memory: 18kB
-> Seq Scan on tab_anti t2_1 (actual rows=100.00 loops=N)
(15 rows)
There shouldn't be Memory mentioned in tests added to the tree.
Greetings,
Andres
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-05-13 13:51 ` Re: Memoize ANTI and SEMI JOIN inner Andres Freund <[email protected]>
@ 2025-05-14 02:13 ` Richard Guo <[email protected]>
0 siblings, 0 replies; 26+ messages in thread
From: Richard Guo @ 2025-05-14 02:13 UTC (permalink / raw)
To: Andres Freund <[email protected]>; +Cc: David Rowley <[email protected]>; Andrei Lepikhov <[email protected]>; pgsql-hackers
On Tue, May 13, 2025 at 10:51 PM Andres Freund <[email protected]> wrote:
> This is failing on CI:
> https://cirrus-ci.com/github/postgresql-cfbot/postgresql/cf%2F5636
> https://api.cirrus-ci.com/v1/artifact/task/5411026402803712/testrun/build-32/testrun/regress/regress...
>
> diff -U3 /tmp/cirrus-ci-build/src/test/regress/expected/memoize.out /tmp/cirrus-ci-build/build-32/testrun/regress/regress/results/memoize.out
> --- /tmp/cirrus-ci-build/src/test/regress/expected/memoize.out 2025-04-12 11:24:10.866868945 +0000
> +++ /tmp/cirrus-ci-build/build-32/testrun/regress/regress/results/memoize.out 2025-04-12 11:32:44.454864476 +0000
> @@ -525,7 +525,7 @@
> -> Unique (actual rows=2.67 loops=N)
> -> Sort (actual rows=67.33 loops=N)
> Sort Key: t2_1.a
> - Sort Method: quicksort Memory: 27kB
> + Sort Method: quicksort Memory: 18kB
> -> Seq Scan on tab_anti t2_1 (actual rows=100.00 loops=N)
> (15 rows)
>
>
> There shouldn't be Memory mentioned in tests added to the tree.
Thanks for pointing this out. You're right — Memory usage should be
hidden from the output of EXPLAIN ANALYZE. I'm a bit surprised the
existing explain_memoize function doesn't already handle that; I
guess it's because the plans in memoize.sql don't currently include
any Sort nodes. In any case, I've added a regexp_replace to filter
out 'Memory: \d+kB' in explain_memoize.
Thanks
Richard
Attachments:
[application/octet-stream] v2-0001-Enable-use-of-Memoize-for-ANTI-joins.patch (9.8K, ../../CAMbWs48g0Q_mWJip_dDEcgaCHZw71MRB=pkz4B6PzWyfVWWaCA@mail.gmail.com/2-v2-0001-Enable-use-of-Memoize-for-ANTI-joins.patch)
download | inline diff:
From 3b1dd3d81b4b7db068d8f7bdf40854c8d0d3edfb Mon Sep 17 00:00:00 2001
From: Richard Guo <[email protected]>
Date: Wed, 9 Apr 2025 18:06:48 +0900
Subject: [PATCH v2] Enable use of Memoize for ANTI joins
Currently, we do not support Memoize for SEMI and ANTI joins because
nested loop SEMI/ANTI joins do not scan the inner relation to
completion, which prevents Memoize from marking the cache entry as
complete. One might argue that we could mark the cache entry as
complete after fetching the first inner tuple, but that would not be
safe: if the first inner tuple and the current outer tuple do not
satisfy the join clauses, a second inner tuple matching the parameters
would find the cache entry already marked as complete.
However, if the inner side is provably unique, this issue doesn't
arise, since there would be no second matching tuple. That said, this
doesn't help in the case of SEMI joins, because a SEMI join with a
provably unique inner side would already have been reduced to an inner
join by reduce_unique_semijoins.
Therefore, in this patch, we check whether the inner relation is
provably unique for ANTI joins and enable the use of Memoize in such
cases.
---
src/backend/optimizer/path/joinpath.c | 47 +++++++++++----------
src/test/regress/expected/memoize.out | 60 +++++++++++++++++++++++++++
src/test/regress/sql/memoize.sql | 27 ++++++++++++
3 files changed, 112 insertions(+), 22 deletions(-)
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 26f0336f1e4..b9e46a40e1f 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -154,13 +154,17 @@ add_paths_to_joinrel(PlannerInfo *root,
/*
* See if the inner relation is provably unique for this outer rel.
*
- * We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
- * matter since the executor can make the equivalent optimization anyway;
- * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, we
- * must be considering a semijoin whose inner side is not provably unique
- * (else reduce_unique_semijoins would've simplified it), so there's no
- * point in calling innerrel_is_unique. However, if the LHS covers all of
- * the semijoin's min_lefthand, then it's appropriate to set inner_unique
+ * We have some special cases: for JOIN_SEMI, it doesn't matter since the
+ * executor can make the equivalent optimization anyway. It also doesn't
+ * help enable use of Memoize, since a semijoin with a provably unique
+ * inner side should have been reduced to an inner join in that case.
+ * Therefore, we need not expend planner cycles on proofs. (For
+ * JOIN_ANTI, although it doesn't help the executor for the same reason,
+ * it can benefit Memoize paths.) For JOIN_UNIQUE_INNER, we must be
+ * considering a semijoin whose inner side is not provably unique (else
+ * reduce_unique_semijoins would've simplified it), so there's no point in
+ * calling innerrel_is_unique. However, if the LHS covers all of the
+ * semijoin's min_lefthand, then it's appropriate to set inner_unique
* because the path produced by create_unique_path will be unique relative
* to the LHS. (If we have an LHS that's only part of the min_lefthand,
* that is *not* true.) For JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid
@@ -169,12 +173,6 @@ add_paths_to_joinrel(PlannerInfo *root,
switch (jointype)
{
case JOIN_SEMI:
- case JOIN_ANTI:
-
- /*
- * XXX it may be worth proving this to allow a Memoize to be
- * considered for Nested Loop Semi/Anti Joins.
- */
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
@@ -715,16 +713,21 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
- *
- * XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
- * = true. Should we? See add_paths_to_joinrel()
+ * Currently we don't do this for SEMI and ANTI joins, because nested loop
+ * SEMI/ANTI joins don't scan the inner node to completion, which means
+ * memoize cannot mark the cache entry as complete. Nor can we mark the
+ * cache entry as complete after fetching the first inner tuple, because
+ * if that tuple and the current outer tuple don't satisfy the join
+ * clauses, a second inner tuple that satisfies the parameters would find
+ * the cache entry already marked as complete. The only exception is when
+ * the inner relation is provably unique, as in that case, there won't be
+ * a second matching tuple and we can safely mark the cache entry as
+ * complete after fetching the first inner tuple. Note that in such
+ * cases, the SEMI join should have been reduced to an inner join by
+ * reduce_unique_semijoins.
*/
- if (!extra->inner_unique && (jointype == JOIN_SEMI ||
- jointype == JOIN_ANTI))
+ if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
+ !extra->inner_unique)
return NULL;
/*
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..150dc1b44cf 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -25,6 +25,7 @@ begin
ln := regexp_replace(ln, 'Heap Fetches: \d+', 'Heap Fetches: N');
ln := regexp_replace(ln, 'loops=\d+', 'loops=N');
ln := regexp_replace(ln, 'Index Searches: \d+', 'Index Searches: N');
+ ln := regexp_replace(ln, 'Memory: \d+kB', 'Memory: NkB');
return next ln;
end loop;
end;
@@ -500,3 +501,62 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+ explain_memoize
+--------------------------------------------------------------------------------------------
+ Aggregate (actual rows=1.00 loops=N)
+ -> Nested Loop Anti Join (actual rows=33.00 loops=N)
+ -> Seq Scan on tab_anti t1 (actual rows=100.00 loops=N)
+ -> Memoize (actual rows=0.67 loops=N)
+ Cache Key: (t1.a + 1), t1.a
+ Cache Mode: binary
+ Hits: 97 Misses: 3 Evictions: Zero Overflows: 0 Memory Usage: NkB
+ -> Subquery Scan on t2 (actual rows=0.67 loops=N)
+ Filter: ((t1.a + 1) = t2.a)
+ Rows Removed by Filter: 2
+ -> Unique (actual rows=2.67 loops=N)
+ -> Sort (actual rows=67.33 loops=N)
+ Sort Key: t2_1.a
+ Sort Method: quicksort Memory: NkB
+ -> Seq Scan on tab_anti t2_1 (actual rows=100.00 loops=N)
+(15 rows)
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+ count
+-------
+ 33
+(1 row)
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+ QUERY PLAN
+-------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on tab_anti t1
+ -> Nested Loop Semi Join
+ Join Filter: (t1.a = t2.a)
+ -> Seq Scan on tab_anti t2
+ -> Subquery Scan on "ANY_subquery"
+ Filter: (t2.b = "ANY_subquery".b)
+ -> Result
+ One-Time Filter: (t2.a > 1)
+ -> Seq Scan on tab_anti t3
+(10 rows)
+
+DROP TABLE tab_anti;
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..8d1cdd6990c 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -26,6 +26,7 @@ begin
ln := regexp_replace(ln, 'Heap Fetches: \d+', 'Heap Fetches: N');
ln := regexp_replace(ln, 'loops=\d+', 'loops=N');
ln := regexp_replace(ln, 'Index Searches: \d+', 'Index Searches: N');
+ ln := regexp_replace(ln, 'Memory: \d+kB', 'Memory: NkB');
return next ln;
end loop;
end;
@@ -244,3 +245,29 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+
+DROP TABLE tab_anti;
--
2.43.0
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-07-02 02:08 ` Richard Guo <[email protected]>
2025-07-02 03:48 ` Re: Memoize ANTI and SEMI JOIN inner wenhui qiu <[email protected]>
1 sibling, 1 reply; 26+ messages in thread
From: Richard Guo @ 2025-07-02 02:08 UTC (permalink / raw)
To: David Rowley <[email protected]>; +Cc: Andrei Lepikhov <[email protected]>; pgsql-hackers
On Thu, Apr 10, 2025 at 7:36 PM Richard Guo <[email protected]> wrote:
> On Wed, Apr 9, 2025 at 6:18 PM David Rowley <[email protected]> wrote:
> > On Wed, 9 Apr 2025 at 18:48, Richard Guo <[email protected]> wrote:
> > > Perhaps we could spend some planner cycles proving inner_unique for
> > > anti joins, so that Memoize nodes can be considered for them?
> > Worth a try. It should be pretty easy to enable, as far as I can see.
> > It might just be a case of shuffling the cases around in the switch
> > statement in add_paths_to_joinrel().
> Right. Here is a patch for that.
Here's a rebased version with no other changes.
David, Andrei, would you like to take another look? I'm planning to
push it soon.
Thanks
Richard
Attachments:
[application/octet-stream] v3-0001-Enable-use-of-Memoize-for-ANTI-joins.patch (9.8K, ../../CAMbWs4_pDZYK+2RGTTjsyoRkRe+Y9rT3Y6b3XtuzcPkNTeP-zw@mail.gmail.com/2-v3-0001-Enable-use-of-Memoize-for-ANTI-joins.patch)
download | inline diff:
From bd3adf290d31151d1939f3505b1846eae1300c79 Mon Sep 17 00:00:00 2001
From: Richard Guo <[email protected]>
Date: Wed, 9 Apr 2025 18:06:48 +0900
Subject: [PATCH v3] Enable use of Memoize for ANTI joins
Currently, we do not support Memoize for SEMI and ANTI joins because
nested loop SEMI/ANTI joins do not scan the inner relation to
completion, which prevents Memoize from marking the cache entry as
complete. One might argue that we could mark the cache entry as
complete after fetching the first inner tuple, but that would not be
safe: if the first inner tuple and the current outer tuple do not
satisfy the join clauses, a second inner tuple matching the parameters
would find the cache entry already marked as complete.
However, if the inner side is provably unique, this issue doesn't
arise, since there would be no second matching tuple. That said, this
doesn't help in the case of SEMI joins, because a SEMI join with a
provably unique inner side would already have been reduced to an inner
join by reduce_unique_semijoins.
Therefore, in this patch, we check whether the inner relation is
provably unique for ANTI joins and enable the use of Memoize in such
cases.
---
src/backend/optimizer/path/joinpath.c | 47 +++++++++++----------
src/test/regress/expected/memoize.out | 60 +++++++++++++++++++++++++++
src/test/regress/sql/memoize.sql | 27 ++++++++++++
3 files changed, 112 insertions(+), 22 deletions(-)
diff --git a/src/backend/optimizer/path/joinpath.c b/src/backend/optimizer/path/joinpath.c
index 7aa8f5d799c..ebedc5574ca 100644
--- a/src/backend/optimizer/path/joinpath.c
+++ b/src/backend/optimizer/path/joinpath.c
@@ -154,13 +154,17 @@ add_paths_to_joinrel(PlannerInfo *root,
/*
* See if the inner relation is provably unique for this outer rel.
*
- * We have some special cases: for JOIN_SEMI and JOIN_ANTI, it doesn't
- * matter since the executor can make the equivalent optimization anyway;
- * we need not expend planner cycles on proofs. For JOIN_UNIQUE_INNER, we
- * must be considering a semijoin whose inner side is not provably unique
- * (else reduce_unique_semijoins would've simplified it), so there's no
- * point in calling innerrel_is_unique. However, if the LHS covers all of
- * the semijoin's min_lefthand, then it's appropriate to set inner_unique
+ * We have some special cases: for JOIN_SEMI, it doesn't matter since the
+ * executor can make the equivalent optimization anyway. It also doesn't
+ * help enable use of Memoize, since a semijoin with a provably unique
+ * inner side should have been reduced to an inner join in that case.
+ * Therefore, we need not expend planner cycles on proofs. (For
+ * JOIN_ANTI, although it doesn't help the executor for the same reason,
+ * it can benefit Memoize paths.) For JOIN_UNIQUE_INNER, we must be
+ * considering a semijoin whose inner side is not provably unique (else
+ * reduce_unique_semijoins would've simplified it), so there's no point in
+ * calling innerrel_is_unique. However, if the LHS covers all of the
+ * semijoin's min_lefthand, then it's appropriate to set inner_unique
* because the path produced by create_unique_path will be unique relative
* to the LHS. (If we have an LHS that's only part of the min_lefthand,
* that is *not* true.) For JOIN_UNIQUE_OUTER, pass JOIN_INNER to avoid
@@ -169,12 +173,6 @@ add_paths_to_joinrel(PlannerInfo *root,
switch (jointype)
{
case JOIN_SEMI:
- case JOIN_ANTI:
-
- /*
- * XXX it may be worth proving this to allow a Memoize to be
- * considered for Nested Loop Semi/Anti Joins.
- */
extra.inner_unique = false; /* well, unproven */
break;
case JOIN_UNIQUE_INNER:
@@ -715,16 +713,21 @@ get_memoize_path(PlannerInfo *root, RelOptInfo *innerrel,
return NULL;
/*
- * Currently we don't do this for SEMI and ANTI joins unless they're
- * marked as inner_unique. This is because nested loop SEMI/ANTI joins
- * don't scan the inner node to completion, which will mean memoize cannot
- * mark the cache entry as complete.
- *
- * XXX Currently we don't attempt to mark SEMI/ANTI joins as inner_unique
- * = true. Should we? See add_paths_to_joinrel()
+ * Currently we don't do this for SEMI and ANTI joins, because nested loop
+ * SEMI/ANTI joins don't scan the inner node to completion, which means
+ * memoize cannot mark the cache entry as complete. Nor can we mark the
+ * cache entry as complete after fetching the first inner tuple, because
+ * if that tuple and the current outer tuple don't satisfy the join
+ * clauses, a second inner tuple that satisfies the parameters would find
+ * the cache entry already marked as complete. The only exception is when
+ * the inner relation is provably unique, as in that case, there won't be
+ * a second matching tuple and we can safely mark the cache entry as
+ * complete after fetching the first inner tuple. Note that in such
+ * cases, the SEMI join should have been reduced to an inner join by
+ * reduce_unique_semijoins.
*/
- if (!extra->inner_unique && (jointype == JOIN_SEMI ||
- jointype == JOIN_ANTI))
+ if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
+ !extra->inner_unique)
return NULL;
/*
diff --git a/src/test/regress/expected/memoize.out b/src/test/regress/expected/memoize.out
index 38dfaf021c9..150dc1b44cf 100644
--- a/src/test/regress/expected/memoize.out
+++ b/src/test/regress/expected/memoize.out
@@ -25,6 +25,7 @@ begin
ln := regexp_replace(ln, 'Heap Fetches: \d+', 'Heap Fetches: N');
ln := regexp_replace(ln, 'loops=\d+', 'loops=N');
ln := regexp_replace(ln, 'Index Searches: \d+', 'Index Searches: N');
+ ln := regexp_replace(ln, 'Memory: \d+kB', 'Memory: NkB');
return next ln;
end loop;
end;
@@ -500,3 +501,62 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+ explain_memoize
+--------------------------------------------------------------------------------------------
+ Aggregate (actual rows=1.00 loops=N)
+ -> Nested Loop Anti Join (actual rows=33.00 loops=N)
+ -> Seq Scan on tab_anti t1 (actual rows=100.00 loops=N)
+ -> Memoize (actual rows=0.67 loops=N)
+ Cache Key: (t1.a + 1), t1.a
+ Cache Mode: binary
+ Hits: 97 Misses: 3 Evictions: Zero Overflows: 0 Memory Usage: NkB
+ -> Subquery Scan on t2 (actual rows=0.67 loops=N)
+ Filter: ((t1.a + 1) = t2.a)
+ Rows Removed by Filter: 2
+ -> Unique (actual rows=2.67 loops=N)
+ -> Sort (actual rows=67.33 loops=N)
+ Sort Key: t2_1.a
+ Sort Method: quicksort Memory: NkB
+ -> Seq Scan on tab_anti t2_1 (actual rows=100.00 loops=N)
+(15 rows)
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+ count
+-------
+ 33
+(1 row)
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+ QUERY PLAN
+-------------------------------------------------
+ Nested Loop Semi Join
+ -> Seq Scan on tab_anti t1
+ -> Nested Loop Semi Join
+ Join Filter: (t1.a = t2.a)
+ -> Seq Scan on tab_anti t2
+ -> Subquery Scan on "ANY_subquery"
+ Filter: (t2.b = "ANY_subquery".b)
+ -> Result
+ One-Time Filter: (t2.a > 1)
+ -> Seq Scan on tab_anti t3
+(10 rows)
+
+DROP TABLE tab_anti;
diff --git a/src/test/regress/sql/memoize.sql b/src/test/regress/sql/memoize.sql
index c0d47fa875a..8d1cdd6990c 100644
--- a/src/test/regress/sql/memoize.sql
+++ b/src/test/regress/sql/memoize.sql
@@ -26,6 +26,7 @@ begin
ln := regexp_replace(ln, 'Heap Fetches: \d+', 'Heap Fetches: N');
ln := regexp_replace(ln, 'loops=\d+', 'loops=N');
ln := regexp_replace(ln, 'Index Searches: \d+', 'Index Searches: N');
+ ln := regexp_replace(ln, 'Memory: \d+kB', 'Memory: NkB');
return next ln;
end loop;
end;
@@ -244,3 +245,29 @@ RESET max_parallel_workers_per_gather;
RESET parallel_tuple_cost;
RESET parallel_setup_cost;
RESET min_parallel_table_scan_size;
+
+-- Ensure memoize works for ANTI joins
+CREATE TABLE tab_anti (a int, b boolean);
+INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
+ANALYZE tab_anti;
+
+-- Ensure we get a Memoize plan for ANTI join
+SELECT explain_memoize('
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;', false);
+
+-- And check we get the expected results.
+SELECT COUNT(*) FROM tab_anti t1 LEFT JOIN
+LATERAL (SELECT DISTINCT ON (a) a, b, t1.a AS x FROM tab_anti t2) t2
+ON t1.a+1 = t2.a
+WHERE t2.a IS NULL;
+
+-- Ensure we do not add memoize node for SEMI join
+EXPLAIN (COSTS OFF)
+SELECT * FROM tab_anti t1 WHERE t1.a IN
+ (SELECT a FROM tab_anti t2 WHERE t2.b IN
+ (SELECT t1.b FROM tab_anti t3 WHERE t2.a > 1 OFFSET 0));
+
+DROP TABLE tab_anti;
--
2.43.0
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-07-02 02:08 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
@ 2025-07-02 03:48 ` wenhui qiu <[email protected]>
2025-07-02 09:01 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
0 siblings, 1 reply; 26+ messages in thread
From: wenhui qiu @ 2025-07-02 03:48 UTC (permalink / raw)
To: Richard Guo <[email protected]>; +Cc: David Rowley <[email protected]>; Andrei Lepikhov <[email protected]>; pgsql-hackers
HI
> - if (!extra->inner_unique && (jointype == JOIN_SEMI ||
> - jointype == JOIN_ANTI))
> + if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
> + !extra->inner_unique)
To be nitpicky, this change is meant to align with the earlier comment
modifications to improve code readability, right? Everything else looks
good to me."
Thanks
On Wed, Jul 2, 2025 at 10:08 AM Richard Guo <[email protected]> wrote:
> On Thu, Apr 10, 2025 at 7:36 PM Richard Guo <[email protected]>
> wrote:
> > On Wed, Apr 9, 2025 at 6:18 PM David Rowley <[email protected]>
> wrote:
> > > On Wed, 9 Apr 2025 at 18:48, Richard Guo <[email protected]>
> wrote:
> > > > Perhaps we could spend some planner cycles proving inner_unique for
> > > > anti joins, so that Memoize nodes can be considered for them?
>
> > > Worth a try. It should be pretty easy to enable, as far as I can see.
> > > It might just be a case of shuffling the cases around in the switch
> > > statement in add_paths_to_joinrel().
>
> > Right. Here is a patch for that.
>
> Here's a rebased version with no other changes.
>
> David, Andrei, would you like to take another look? I'm planning to
> push it soon.
>
> Thanks
> Richard
>
^ permalink raw reply [nested|flat] 26+ messages in thread
* Re: Memoize ANTI and SEMI JOIN inner
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-07-02 02:08 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-07-02 03:48 ` Re: Memoize ANTI and SEMI JOIN inner wenhui qiu <[email protected]>
@ 2025-07-02 09:01 ` Andrei Lepikhov <[email protected]>
0 siblings, 0 replies; 26+ messages in thread
From: Andrei Lepikhov @ 2025-07-02 09:01 UTC (permalink / raw)
To: wenhui qiu <[email protected]>; Richard Guo <[email protected]>; +Cc: David Rowley <[email protected]>; pgsql-hackers
On 2/7/2025 05:48, wenhui qiu wrote:
> HI
>
> > - if (!extra->inner_unique && (jointype == JOIN_SEMI ||
> > - jointype == JOIN_ANTI))
> > + if ((jointype == JOIN_SEMI || jointype == JOIN_ANTI) &&
> > + !extra->inner_unique)
> To be nitpicky, this change is meant to align with the earlier comment
> modifications to improve code readability, right? Everything else looks
> good to me."
Yep, I also found only this flaw.
Comments looks clear to me, test is quite stable.
May be correct the line:
INSERT INTO tab_anti SELECT i%3, false FROM generate_series(1,100)i;
with a backspace or an 'AS' keyword?
--
regards, Andrei Lepikhov
^ permalink raw reply [nested|flat] 26+ messages in thread
end of thread, other threads:[~2025-07-02 09:01 UTC | newest]
Thread overview: 26+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2020-12-03 15:19 [PATCH 3/3] Extended statistics on expressions Tomas Vondra <[email protected]>
2025-03-06 13:08 Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-19 17:15 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-20 06:02 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-21 15:56 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 02:33 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 02:50 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:04 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 03:21 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 03:33 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 06:34 ` Re: Memoize ANTI and SEMI JOIN inner Alena Rybakina <[email protected]>
2025-03-31 07:45 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 09:03 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:39 ` Re: Memoize ANTI and SEMI JOIN inner David Rowley <[email protected]>
2025-03-31 09:50 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 09:46 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-03-31 10:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-03-31 10:33 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[email protected]>
2025-04-01 07:18 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-09 06:48 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-04-10 10:36 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-05-13 13:51 ` Re: Memoize ANTI and SEMI JOIN inner Andres Freund <[email protected]>
2025-05-14 02:13 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-07-02 02:08 ` Re: Memoize ANTI and SEMI JOIN inner Richard Guo <[email protected]>
2025-07-02 03:48 ` Re: Memoize ANTI and SEMI JOIN inner wenhui qiu <[email protected]>
2025-07-02 09:01 ` Re: Memoize ANTI and SEMI JOIN inner Andrei Lepikhov <[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